# Reserve Banks, the Discount Rate Recommendation, and FOMC Policy.

Geoffrey M. B. Tootell [*]

In the United States, private citizens play little direct role in policymaking. The directors of the boards of the regional Federal Reserve banks are an apparent exception to this rule. These directors recommend changes in the discount rate, although the Board of Governors decides whether to act on the recommendations. These directors would have greater influence if they affected the FOMC votes of their district bank presidents. This paper shows that the FOMC votes of the regional bank presidents are strongly correlated with the discount rate recommendation of their bank's board. Several alternative explanations for the correlation are then examined.

1. Introduction

The directors who sit on the boards of the 12 district Federal Reserve banks are the rare private citizens who have a direct role in government decision making. These nine business, labor, financial, and community leaders are neither elected nor appointed by elected officials; as a result, their goals could differ from those of the nation as a whole. Although these directors do not actually set policy, they do recommend changes in the discount rate to the Board of Governors of the Federal Reserve System. Thus, on the surface their role in monetary policy deliberations is only advisory. However, their influence could in fact be substantially more important if they actually affected the Federal Open Market Committee (FOMC) votes of their respective Reserve bank presidents. This paper examines this more significant link to monetary policy by testing the relationship between the discount rate recommendation of the Reserve bank's board of directors and the vote of that district's bank president at the FOMC. It is shown that the FOMC votes of Reserve bank presidents are significantly correlated with their board's current discount rate recommendation.

Given the distinct responsibilities of the district board and the bank president, it is entirely possible that the president's vote would differ from the discount rate recommendation. By law, each Reserve bank's board of directors must make a discount rate recommendation to the Board of Governors every two weeks. On the other hand, every eight weeks a Reserve bank president must decide on his or her own view of the proper course for monetary policy at the FOMC meeting. De jure, there is no reason that the two decisions should conform. De facto, it is unclear whether, or by how much, the local board's discount rate recommendation influences the president's FOMC deliberations or whether or by how much the district president influences the local board's discount rate recommendation. This paper presents evidence that the two decisions are made separately but that the discount rate recommendation does help explain the president's FOMC vote.

For several reasons, one might expect the determinants of the local board's discount rate recommendation to differ from those of the president's FOMC vote. The local boards may have concerns and perspectives that deviate from those of FOMC members in general and their own president in particular. For example, members of the district board may disproportionately weight their region's economic performance, while the bank's president may have a stronger national focus. In fact, McNees (1993) shows that several district boards respond to regional economic conditions, while Tootell (1991) finds that district bank presidents do not base their FOMC votes on the economic performance of their regions. It is also possible that the district banks' boards of directors have different national goals than members of the FOMC; for example, local boards may disagree with their president on the relative costs of inflation and unemployment. Finally, the district board's forecast could systematically diverge from that of its pr esident. Any one of these possible causes could produce a schism between the discount rate recommendation of the bank's board of directors and the FOMC votes of its president.

On the other hand, there are several reasons why the two decisions should be highly correlated. As a national instrument, the FOMC vote of a bank president should depend on national economic conditions. The performance of the economy should be highly correlated with the variables, be they national, regional, or industry specific, that determine the local boards' discount rate recommendations. It is shown, however, that the macrovariables thought to be significant determinants of FOMC behavior do not explain all the correlation between a Reserve bank's discount rate recommendation and its president's FOMC vote. Alternatively, a bank's discount rate recommendation and its president's FOMC vote might be correlated because either the local board or the president is, essentially, making both decisions. However, evidence presented in this paper suggests that neither the district bank's board nor its president dominates both decisions. Finally, the district presidents and their boards may tend to agree on monetary policy and disagree with the rest of the Federal Reserve System because they share an outlook for the economy that differs from the rest of the FOMC, because they share goals at variance with the rest of the FOMC, or because they use a different model than the FOMC to analyze the data. All three of these explanations are explored in this paper.

The next section briefly describes the data. Section 3 then examines the correlation between the local board's discount rate recommendation and the bank president's FOMC vote. The discount rate recommendation is shown to add information to the explanation of the bank president's FOMC vote even when the relevant national and regional data are included. No obvious omitted macrovariable accounts for this result. Section 4 examines the possible sources of the discount rate recommendation's effect on the presidents' FOMC votes. The evidence suggests that the district bank's board of directors does influence, but does not control, the FOMC vote of the bank's president. Section 5 concludes.

2. The Data

The FOMC votes of district presidents and the discount rate recommendations of the regional boards are merged with data describing the current and expected future state of the economy. The FOMC votes were recovered from the minutes of the FOMC meetings, published approximately six weeks after each gathering. These minutes, and particularly the directive, outline the policy adopted and the vote of each FOMC member on that action. The wording of the directive has evolved over time; thus, at certain times it was easier to glean the course of monetary policy from the directive than at others. [1] Once the course of monetary policy is translated from the directive, judging the monetary policy action voted by each FOMC member is straightforward. When a member dissents from the chosen policy, the dissent is explained and the desired policy articulated. At any one meeting, seven governors and five presidents are permitted to vote, so in general 12 votes are registered. Finally, it should be emphasized that the direc tive provides only the direction of policy--either to tighten policy, loosen it, or keep it constant--and does not give the magnitude of any policy change. As a result, the dependent variable is discrete and trichotomous.

The data on the discount rate recommendation for each bank back to 1974 were retrieved from the discount rate minutes of the Federal Reserve. Every week the Board of Governors meets to consider action on the discount rate. If a recommendation is in, and it is in the direction that the Board of Governors desires, the members can vote to alter the rate. The minutes of that meeting provide information on any changes in the discount rate and on which banks recommended what change at the time of the meeting.

Several problems arise when merging data on each bank president's monetary policy vote with his or her board's discount rate recommendation. The district bank's discount rate recommendation is updated every two weeks. The FOMC, on the other hand, met roughly once a month prior to 1980 and has met eight times a year since. The higher frequency of the discount rate series must be altered to conform to the FOMC frequency. Since this paper examines the effect of a bank's discount rate recommendation on the bank president's FOMC vote, only the recommendation in at the time of the FOMC meeting is analyzed. Furthermore, discount rate recommendations made during periods when the district bank president was not a voting member of the FOMC cannot be used to examine the effect of these recommendations on the president's FOMC vote. Even though a discount rate recommendation is in, the district president's reaction to that recommendation can be quantified only when he or she votes at the FOMC.

The discount rate recommendations and the FOMC votes should, in general, be highly correlated since the Reserve banks' boards of directors and the members of the FOMC are reacting, certainly in part, to the same economy. Thus, when examining the effect of any variable on the monetary policy votes of FOMC members or on the local boards' discount rate recommendations, it is necessary to account for the state of the economy at the time these actions were taken. Since monetary policy works with long lags, changes in monetary policy should be determined by expectations about the future state of the economy. The expectations of the district presidents and their boards are instrumented for using the economic forecasts of the Federal Reserve Board staff, and these are circulated immediately prior to every FOMC meeting in the "Green Book." [2] Contained in the Green Book are Reserve Board staff forecasts of real gross national product (GNP) and its components as well as the unemployment rate and several measures of i nflation. Since the discount rate minutes are spotty before 1974 and the Green Book forecasts become available to the public only with a five-year lag, the sample covers the years 1974 to 1993.

McNees (1986, 1992) and Tootell (1991, 1997) show that the forecasted variables that affect monetary policy include the growth of real GNP, the unemployment rate, and the rate of inflation. Lagged values of the target variables are also examined to ensure that the results are not an artifact of imposing forward-looking behavior on all the policymakers. Including these macrovariables in the analysis ensures that any remaining correlation between the FOMC votes of the district presidents and the discount rate recommendations of their local boards is independent of the current and expected state of the economy.

3. Discount Rate Recommendations and FOMC Votes

All 12 Reserve banks are pooled in the analysis, although this constraint is relaxed later in the paper. The discount rate recommendation is also transformed into a trichotomous discrete variable equal to --1 if the local board advocates a reduction in the discount rate, zero if it advocates no change in the rate, and 1 if it advocates an increase in the rate. [3]

Because the dependent variable takes on only three discrete values, least squares estimation would produce inefficient estimates of the coefficients and inconsistent estimates of their standard errors. For this reason, the presidents' voting function is estimated as a multinomial logit. Although multinomial logit estimation of the coefficients and their standard errors produces efficient and consistent estimates, the coefficients are not as easily interpreted as those produced by least squares. The nonlinearity of the logit function results in the estimated effect of a variable on the FOMC members' voting probabilities depending on the values of the other right-hand-side variables. Thus, the economic significance of any variable must be measured at a given value of the other explanatory variables, usually selected to be their means. Finally, in order to identify the parameters in the model, the coefficients on the probability to keep policy unchanged are set equal to zero; as a result, the equations estimate the probability of voting to tighten or loosen policy relative to a vote of no change.

The coefficients from a multinomial logit estimation of the Reserve bank presidents' FOMC votes on a constant and their respective boards' discount rate recommendations are presented in the first column of Table 1. The top panel presents the coefficients measuring the effects of the explanatory variables on the probability of the presidents voting for a tightening in policy relative to the probability of voting for no change, while the bottom panel presents the coefficients estimating the effects of the independent variables on the probability of the presidents voting for a loosening in policy relative to the probability of voting for no change. In column 1, the coefficients on the discount rate recommendation reveal that presidents' votes at the FOMC are highly correlated with their bank's discount rate recommendation in at the time of the FOMC meeting. A bank president is much more likely to vote for a tightening and much less likely to vote for a loosening the more restrictive is the local board's discount rate recommendation. As mentioned earlier, this result should come as no surprise since the local boards of directors and the district presidents tend to react to the same economic variables; it is probable that similar expectations about the same macrotargets are driving both the discount rate recommendation and the FOMC vote.

To account for these expectations, column 2 of Table 1 assumes that the district presidents target some form of nominal GNP; policy depends on the expected growth in real GNP and the forecast of inflation. These forecast variables have the expected effect on the FOMC votes of the presidents. High expected inflation significantly increases the probability of tightening, and rapid real output growth significantly lowers the probability of loosening. Yet even after controlling for the expected state of the economy, the discount rate recommendation remains a significant determinant of the president's FOMC vote. If the local board's discount rate recommendation and the bank president's FOMC vote were merely reacting independently to the forecasted growth in prices and real output, then the discount rate recommendation should add no information to an equation explaining the FOMC votes that included the forecasts of these two variables. Inclusion of these forecasts, however, has almost no effect on the estimated pa rtial correlation between the discount rate recommendation and the district president's FOMC vote; the local board's discount rate recommendation is still statistically and economically significant in predicting the district president's vote at the FOMC.

The significance of the discount rate recommendation does not seem to arise from a simple misspecification of the presidents' voting function. In column 3, the coefficients from a less parsimonious model only reinforce the conclusion that the discount rate recommendation of each bank's board is significantly related to the FOMC vote of that bank's president. Money growth and the last known unemployment rate are added to the simple nominal GNP targeting rule. Including a monetary aggregate in the voting function makes sense if some district presidents are monetarists or if some district presidents believe that past money growth improves the Green Book's forecast of inflation. The unemployment rate should be included if it, rather than real GNP growth, is in the bank presidents' objective function. [4] The coefficients on the economic variables in column 3 are, for the most part, significant and correctly signed. In general, higher expected real growth, higher forecasted inflation, more rapid money growth, and lower expected unemployment increase the probability that the FOMC will vote to tighten policy and decrease the probability that it will vote to loosen policy. Yet the coefficients on the discount rate recommendation remain large and statistically significant. Adding these other variables to the FOMC voting function seems to have little effect on the size or significance of the coefficient on the discount rate recommendation.

The equations presented in Table 1 are clearly not exhaustive, and a misspecification could still be causing the significance of the discount rate recommendation. It is possible that the FOMC has a forecast horizon different from the one modeled in Table 1; the committee could be more forward or more backward looking. The significance of the discount rate recommendation in Table 1 may simply be due to a misspecification of the horizon to which the FOMC reacts. It is also still possible that some other macrovariable may be missing from the analysis. Both of these hypotheses will be examined in turn.

Different Forecast Horizon

The inclusion of only the one-quarter-ahead forecast may miss an important dynamic element of the economy that both the FOMC and the local board of directors are considering. Policymakers care about the economy's momentum and its performance over a longer time period than one quarter. One way to capture this longer-run performance is to include either lagged data or longer-horizon forecasts.

Table 2 explores the effects of altering the forecast horizon in the voting function. The specification in the first two equations is similar to that in column 3 of Table 1, except now explicit forward-looking data on the unemployment rate and backward-looking data on the core inflation rate are included. The first column of Table 2 provides some evidence that Reserve bank presidents are partially backward looking on core inflation. Lagged inflation has a significant and correctly signed effect on the probability of loosening but an incorrectly signed effect on the probability of tightening. However, the collinearity of the lagged and forecasted inflation measures makes it difficult to distinguish the effects of backward- and forward-looking inflation. [5]

To ensure that the significance of the discount rate recommendation in the equations in Table 1 is not accounted for by the omission of lagged data, Equation 2 adds the discount rate recommendation to an equation that includes the backward-looking variables. The coefficients on the discount rate recommendation of the bank's board are still large and statistically significant well beyond the 1% level. The omission of the lagged variables does not explain the significance of the discount rate recommendation.

Altering the specification to include forecasts as far as one year ahead also fails to diminish the importance of the discount rate recommendation on the presidents' FOMC vote. In column 3, the presidents' voting function is reestimated, including the forecasts of real GNP growth and inflation over the next six months and the six months after that, along with the backward-looking variables. As discussed in Tootell (1997), this specification smoothes out the quarterly noise in the forecast while examining the reaction to longer-term forecasts. The coefficient estimates presented in column 3 reveal that the longer-run forecasts are important, but the inclusion of all these variables has little effect on the strength of the relationship between the local board's discount rate recommendation and that bank president's FOMC vote. The importance of the discount rate recommendation to the president's FOMC vote does not appear to be an artifact of the forecast horizon examined in these voting functions.

Omitted Macrovariables

Although the forecasts should incorporate all the relevant macroinformation, and the ultimate goals of monetary policy seem to be relatively few and agreed on, some omitted macro-variable may still be causing the discount rate recommendations and the presidents' FOMC votes to move together. One approach to testing for omitted macrovariables is to examine the effects on Fed policy of a wide range of potential candidates, as is done in McNees (1986, 1992). Since the final specification in McNees (1992) looks very much like the specifications examined here, this approach will not be pursued. As an alternative, the coincidence of discount rate recommendations across districts is examined. If serious omitted macrovariables exist, agreement on their macroeffects should be widespread, and the discount rate recommendations should be highly correlated both across district banks and with actual FOMC policy changes.

In fact, most recommendations for discount rate changes were advocated without widespread agreement across the different Reserve bank boards of directors or across a majority of the FOMC. Over 50% of all recommended changes in the discount rate were not followed by an FOMC policy change in the recommended direction. When the discount rate recommendation did forecast FOMC policy changes, much of the coincidence could be accounted for by the current and expected state of the economy captured by the economic variables examined in the previous tables. Furthermore, when at least one district bank had a recommendation in to change the discount rate, the average number of banks with a recommendation in for change was only three. A majority of the time, the FOMC policy overall did not appear to agree with the recommended change in the discount rate, nor did the other banks' boards, suggesting that a pressing omitted macroevent probably does not explain the effect of the discount rate recommendation on the FOMC vote.

As mentioned previously, this low raw correlation of the recommendations for changes in the discount rate across district banks fails to take into account the state of the economy. If variables omitted from the estimation of the voting function were driving every Reserve bank's recommendation in a similar direction, then the correlation across banks of the residual from an estimation of the discount rate recommendation on the included macrovariables should be high. The average correlation across Reserve banks of this residual is 0.33, and the average multiple correlation of one bank's residuals with all banks is 0.42. [6] The relationship is positive, as would be expected if omitted macrovariables explained the discount rate's importance in the voting equation. However, the correlation is not very high. In fact, most of the times when more than one Reserve bank was in with a discount rate recommendation, the macroforecasts would have predicted such a recommendation; the variables in the estimation account fo r much of any correlation seen across the different Reserve banks' discount rate recommendations. Most of the movement of the discount rate recommendation unaccounted for by the economic data in these voting functions is not explained by the behavior of the other Reserve banks' residuals; it appears to be largely idiosyncratic, not shared.

And, what is also important, the district presidents responded to the idiosyncratic recommendations. The bank president's FOMC vote did not react to the discount rate recommendation only when the rest of the FOMC concurred with the recommendation. Bank presidents dissented about 20% of the time when their board had a recommendation in for a change and the FOMC did not change policy in that direction; this is a much higher frequency than the average dissent rate of 6%. In fact, almost 50% of all dissents by presidents occurred when their board had a discount rate request in for a policy change different from the policy action taken by the FOMC. The importance of the discount rate recommendation, even when the FOMC disagreed with it, suggests that the results are not driven by an omitted macrovariable.

The correlation of the local board's discount rate recommendation with its president's FOMC vote does not seem to be due to macrovariables excluded from the analysis. Not only should the forecasts capture most relevant macroevents, but neither the discount rate recommendations nor the unexplained part of the recommendations is highly correlated across districts. It appears that some shared perspective of the local board and the district president is affecting the FOMC votes of the bank presidents. The next section examines the possible nature of this shared perspective.

4. The Nature of the Idiosyncratic Information

Although the local board's discount rate recommendation helps predict its president's FOMC vote, the nature of the link between the two policy actions is not immediately clear. Certainly there is no necessary correspondence between the two, as the regional boards can recommend any discount rate they want and the district presidents can vote any way they wish at FOMC. However, several explanations for the correlation are possible. First, the local board's recommendation may directly affect the president's FOMC vote. [7] Alternatively, the reverse could be true; the president could be influencing the discount rate recommendation of the local board. In this case, the discount rate recommendation might reveal the president's intention before he or she gets to the FOMC. Finally, the correlation may be picking up a joint effect on the outlook of both the president and the local board, that may differ from the outlook of the rest of the FOMC; for example, a district president's forecast of, goals for, or model of t he economy may have common elements with those of his or her board of directors that are not shared by the rest of the FOMC. Any of these explanations could be responsible for the correlation between the discount rate recommendation and the district president's FOMC vote. Which of the above three explanations is more likely is examined here.

Are the Actions of the District Boards and Presidents Independent?

If the district board reacts differently to the economy when determining its recommendation for the discount rate than the district president does when formulating his or her FOMC vote, then estimates of the presidents' voting function should differ from estimates of the local board of directors' discount rate recommendation function. Conversely, if either the president or the local board decides both the discount rate recommendation and the FOMC vote, then the determinants of both decisions should be identical. [8]

To examine the similarity between these two decisions, the discount rate recommendation must be estimated as a function of the variables used to explain the presidents' FOMC votes. When the discount rate recommendation is a dependent variable, it is constructed to be comparable to that of the FOMC vote. Thus, in column 2 of Table 3, the top panel provides the estimated effects of the macrovariables on the local boards' probability of recommending an increase in the discount rate (a tightening) relative to a recommendation of no change, and the bottom panel gives the coefficients for the same variables' effects on the local board's probability of recommending a decrease in the discount rate (a loosening) relative to a recommendation of no change in that rate. The forecasts are used as instruments of the local board's expectations.

The first equation in Table 3 estimates the monetary policy stance of both the presidents and the local boards when the coefficients on the determinants of both the discount rate recommendation and the FOMC vote are constrained to be identical, which is the proper specification if the same person or committee made both decisions. The dependent variable in column 1 is the monetary policy vote of either the presidents at the FOMC or the desired discount rate move of the local board. Thus, the estimates presented in column 1 assume that the vote of a president for tighter or looser policy is determined by the same forces as the recommendation of the local board for a higher or lower discount rate. Columns 2 and 3 present the coefficients from the unconstrained estimation, where the coefficients on the presidents' FOMC votes are allowed to differ from those of the local boards' discount rate recommendation. The significance levels of the coefficients in columns 2 and 3 reveal that the two decisions depend on man y of the same variables. Yet it can be rejected well beyond the 1% level that the two sets of coefficients are the same. The results suggest that the two decisions are made differently, supporting the hypothesis that neither the bank president nor the district board of directors dominates both decisions.

However, it is possible that the discount rate recommendation and the federal funds rate serve different functions. Differences in the functions of the two instruments could explain differences in their estimated determinants. A fairly extensive literature explores the different roles that the discount rate and the federal funds rate may play. For example, Thornton (1992) and Rush, Sellon, and Zhu (1994) have examined the announcement value of discount rate changes, suggesting that the discount rate is used to signal a more permanent change in interest rates, while the federal funds rate is used to probe changes.

In fact, the evidence provides little support for the conclusion that the discount rate recommendation is used as an announcement device. A greater reluctance to use the discount rate should result in a different constant term in the discount rate recommendation equation compared to the FOMC voting function, but it should not produce different coefficients on the economic variables in these two functions. In fact, when the constant term is allowed to differ across the two decisions, the coefficients on the economic variables still differ significantly across the two decisions. On a more intuitive level, if the discount rate recommendation was meant to signal a long-run change in rates and the same person was making both decisions, then we should frequently see bank presidents voting to change the federal funds rate and not the discount rate but rarely see the opposite. In fact, 50% of the time that a recommendation for a change in the discount rate was in, it was not ratified by a change in FOMC policy, nor did the district president oppose the inaction--the opposite of the direction to be expected according to the announcement effect.

Alternatively, it is possible tat the discount rate recommendation represents the bank president's inclination for monetary policy, even though he or she may be unwilling to dissent at the FOMC. If the district president were reluctant to dissent but wanted to reveal policy inclinations in a more subtle way, one would expect to see many more recommendations for discount rate changes than FOMC votes for change. Conversely, one would not expect to see FOMC votes to change policy without a discount rate recommendation in. Yet over 60% of FOMC dissents occurred without a similar discount rate recommendation in from that president's local board. In fact, the district bank frequently had a recommendation in for a policy move in one direction while its president voted for a move in the opposite direction. The data are not consistent with the hypothesis that the president or the local board of directors made both decisions.

Finally, it must be emphasized that the previous literature examines the actions by the Board of Governors in setting the discount rate, not the desires of the regional boards of directors in recommending changes in this rate. Recommendations about the discount rate are not the same as changes in the discount rate. Even if the discount and federal funds rates serve different functions, it is not clear that the district boards of directors feel constrained by these differences. Local boards are likely to consider their recommendations as a signal for their desired federal funds rate changes. Since the Board of Governors actually controls the discount rate, the constraints caused by any nuances of the potentially different functions that the two instruments may serve are left for the Board of Governors to sort out. Since the funds rate, not the discount rate, directly affects the economy, it makes sense that the discount rate recommendation is used by the regional board of directors as a signal of its desired funds rate changes.

In an attempt to uncover the source of the different policy reactions between the presidents and their local boards, column 4 of Table 3 presents coefficient estimates of variables that interact a dummy variable indicating whether the monetary policy decision being made is a discount rate recommendation with all the other explanatory variables. Note that this sample is the same as that used in column 1 since the null hypothesis is that the two sets of coefficients are identical. In general, it appears that the monetary policy desired by the local boards of directors may be more activist than the policy voted on by the district bank presidents. The local boards seem to react much more strongly to the real economy, as the coefficients on the lagged unemployment rate and the expected growth in real output are larger in absolute terms. Some evidence also suggests that the local boards react more strongly to inflation. Again, if the president or the local boards simply controlled these two different instruments, it is not clear why these particular variables should explain their different uses. The results in Table 3 demonstrate that presidents at the FOMC respond differently to the state of the economy than do their local boards, suggesting that the discount rate recommendation is not simply foreshadowing the president's desires, nor is the president simply following the local board's orders.

Regional Information

Although the two decisions appear to be somewhat independent, the marginal predictive power of the discount rate recommendation on the president's FOMC vote suggests significant interrelation. The policy desires of the president and the local board might coincide while differing with the FOMC as a whole for several reasons. The most obvious explanation is that they both give more weight to their district's economic performance than its share in GDP. Tootell (1991) examined the role played by regional economic indicators in the FOMC voting of district presidents and found that regional economic conditions, as measured by employment growth and the unemployment rate in the district, added no marginal explanatory power to the national data. McNees (1993), however, found that regional economic indicators did affect the discount rate recommendation of some local boards. Some evidence suggests that region-specific economic information affects the votes of some local boards, but there is little evidence that these e ffects spill over into the FOMC votes of the presidents.

Table 4 merges these two findings by examining the effect of regional employment growth and unemployment rates on the formation of monetary policy. Column 1 presents the estimates of the effect of the Reserve district's employment growth on the local board's discount rate recommendation. Regional employment growth is not significant. [9] The second column shows the results when the district's unemployment rate, rather than its employment growth rate, is used as a proxy for the region's economic health. A higher regional unemployment rate, holding the national rate constant, does increase the probability that the local boards of directors recommend a lowering of the discount rate. The coefficient on the regional unemployment rate provides some evidence that local boards determine their discount rate recommendations on the basis of how well the regional economy is doing.

The results from an examination of the effect of the local economy on the district president's FOMC vote are presented in the next two columns. Neither the regional employment growth rate nor the unemployment rate is a significant determinant of how the district presidents will vote at the FOMC; the evidence suggests that the district presidents pay attention to the macrodata rather than the regional data. These results are robust to different specifications and consistent with the findings in Tootell (1991) and McNees (1993). [10] More important for this study, column 5 of Table 4 shows that the partial correlation of the discount rate recommendation with the FOMC vote of that district bank's president is unaffected by the inclusion in the analysis of the region's economic performance. The correlation between the discount rate recommendation and the president's FOMC vote is not due to shared concerns about the economic performance of the region.

The difference in the use of regional information across the two policy decisions also supports the conclusion that the two decisions are made by different groups. If the two decisions were made by the same people, there could be some reason to suspect that the constant term in the discount rate recommendation equation would differ from that in an FOMC voting equation; there would be little reason to expect the coefficients on the macroforecast variables to differ and no reason to expect that regional information should be important for the discount rate but not for the federal funds rate. While regional information does not explain the correlation between the discount rate recommendation and the presidents' FOMC vote, it does emphasize the difference in the objective function of the two groups making the decisions.

Models, Goals, and a Regional Outlook

The two decisions appear to be fairly independent, and although the correlation is not explained by region-specific economic conditions, regional information in a broader sense may be its origin. One possibility is that individual Reserve banks may have models of the economy or goals for monetary policy that differ from the rest of the FOMC. For example, the Federal Reserve bank of St. Louis has historically had a monetarist outlook, while the Federal Reserve bank of Cleveland tends to have a strong distaste for inflation (Hoskins 1991). Since the local board of directors plays an important role in the selection of the Reserve bank president, agreement between the local board and the president about the economic model and the appropriate goals for monetary policy may explain why their monetary policy reactions are partially correlated yet can be somewhat different from those of the other FOMC members.

To capture the effect of potential differences in goals, individual Reserve bank dummy variables were included in the estimation. Since the constant term is a linear combination of the coefficient estimates and these goals, estimating the logits with different regional bank constant terms permits the goals to differ between banks. Allowing the constants to vary across the district banks has no effect on the results; the discount rate recommendation of the local board continues to have a strong effect on the FOMC vote of the Reserve bank's president. To capture the possible effects of variation across banks in the model used, the most distinctive model difference is also examined--whether the bank is monetarist. The coefficient on money is allowed to differ for the three banks with the most consistent monetarist outlook. [11] Allowing this coefficient to vary fails to account for the size or significance of the coefficient on the discount rate recommendation. Differences in the banks' goals or models do not s eem to be the answer.

It is also possible that the outlook for the national economy could depend on the region. The national forecast could vary because of region-specific experience not associated directly with its economic performance. In this case, the Green Book forecasts would be missing the shadings of the outlook that depend on the region in which the policymaker lives. The credit crunch in New England in the early 1990s, as examined by Peek and Rosengren (1995), is one possible example of how the national outlook can depend on the region. In the early 1990s, a credit squeeze in New England had the potential to expand to the rest of the country since any district where real estate values were weak was subject to the shock. It is possible that the assessment of the probabilities that this effect would spread, and the assessment of the consequences if it did, may have been more dire in New England. Thus, at that time, both the local board and the bank president at the FOMC may have been more negative about the national econo my than were the Green Book forecasts and the rest of the FOMC.

As seen in the previous section, distinguishing the exact source of the differences in the shadings of the outlook is difficult and beyond the scope of this paper. However, one way to attempt to capture the regional outlook independent of the region's economic performance is to examine whether members of the FOMC find another region's or another bank's shading on the forecast compelling. Sticking to the credit crunch example, district presidents from areas where real estate values were beginning to fall may have found reservations about the economy based on the possible effect of the credit constraints more sympathetic. In this way, idiosyncratic differences about the national outlook could influence other FOMC members. This influence is measured in Table 5. The effect of each bank's discount rate recommendation on the FOMC vote of the other FOMC members, when the state of the economy and the discount rate recommendation of that bank are also taken into account, is shown. For brevity, Table 5 presents only t he coefficients measuring the effect of each bank's discount rate recommendation on all the other FOMC members.

The recommendations of local boards often do affect the other members of the FOMC, as would be true if the banks brought information to the meeting that the other FOMC members found persuasive. Ten of 24 such coefficients are correctly signed and statistically significant beyond the 10% level. As an example, note that the discount rate recommendation from New York significantly affects the probability of both tightening and loosening by the other FOMC members. New York's effect could be due to capital market information that many at the FOMC found important to the shading of the national outlook. The different outlooks or information could be the source of the partial correlation between the board's discount rate recommendation and the president's FOMC vote--their persuasiveness at the FOMC suggests the potential value of these different outlooks to the formation of monetary policy. [12]

5. Conclusion

A district's discount rate recommendation is clearly a powerful signal of that Reserve bank president's FOMC vote. Although the discount rate recommendation does depend on many of the same economic variables as the FOMC vote, it adds marginal predictive power for the FOMC vote beyond these macrovariables. Yet the discount rate recommendation is not simply the district president telegraphing his or her punch, and the FOMC vote is not the president simply following orders from his or her local board since the determinants of the two decisions differ significantly. The correlation seems to derive from idiosyncratic information not related to the macroforecasts in the Green Book but shared at the regional level. Different information appears to be emphasized in the different regions, and some evidence suggests that the FOMC as a whole finds this information useful.

The implications for monetary policy are important. If monetary policy were being disproportionately influenced by concerns about the regional economic performance or by narrow private interests, the federal structure could be suboptimal. There is little evidence that local conditions in these regions overly affect presidents' votes at the FOMC meetings. However, the meeting of the local boards and district presidents does seem to affect the presidents' FOMC votes, and this could be beneficial if it results in the presentation of different information, different points of view, or different forecasts at the FOMC meeting. In fact, there is evidence that the discount rate recommendations related to idiosyncratic information at the district level often affect the votes of the other FOMC members; this highlights one important benefit of the Federal Reserve System--its inclusion of different views for a full debate. In this way, the presentation of different perspectives at the FOMC justifies its structure as a c ommittee including representation from different regions.

(*.) Research Department, Federal Reserve Bank of Boston, 600 Atlantic Avenue, Boston, MA 02106-2076, USA; E-mail geoff.tootell@bos.frb.org.

The author wishes to thank Lynn Browne, Richard Kopcke, Stephen McNees, and two anonymous referees for their helpful comments. Valuable research assistance was provided by Rokeya Kahn and Peter Morrow, The analysis and conclusions of this paper are not necessarily endorsed by the Federal Reserve System or the Federal Reserve Bank of Boston.

Received December 1997; accepted August 1999.

(1.) Since 1997, the FOMC has released the explicit target for the federal funds rate after each meeting. Earlier, the directive called for maintaining, tightening, or easing either "reserve conditions" or the federal funds rate. The assessment of monetary policy from the directive depended on which of these alternatives was called for by the directive. Only during the monetary experiment of the early 1980s did judgment play an Important role in discerning the course of monetary policy.

(2.) The local boards of directors do not have access to the Green Book forecasts. Nonetheless, the Green Book projections are used as instruments for more widely available market forecasts. The results are similar when publicly available forecasts, suhs as those from DRI, are used. Any potential informational advantage contained in the Green Book would bias the results toward rejecting that the local board's discount rate recommendation had a partial correlation with the president's FOMC vote.

(3.) The discount rate recommendation is transformed this way for several reasons. First, the recommendations tend to be fairly discontinuous; they take on values in 25-basis-point multiples from 100 to --100. Generally, the recommendations are for 50-basis-point movements, so constraining the discount rate recommendation to take on one of only three values is not that restrictive. Also, it will be useful later to have the discount rate recommendation and the FOMC vote in comparable form. Finally, when separate dummies representing times when a Reserve bank had a recommendation in for a discount rate tightening and times when it had a recommendation in for a discount rate loosening were included in the analysis, the results were identical, and relaxing the assumption of a symmetric response did not increase the explanatory power of the estimation. As defined, as the variable increases, the probability of the president voting for a tightening should rise, and his or her probability of voting for a loosening sh ould fall.

(4.) If Okun's law holds with regularity, giving today's unemployment rate and the forecasted growth in gross domestic product (GDP) is equivalent to giving the forecast of the unemployment rate.

(5.)The pattern of the coefficients on inflation for the probability of loosening suggests that changes in inflation may be Important; however, there is little evidence of such a relationship for the probability of tightening. The estimation in the second and third columns of Table 2 allows the specification to differ for the two different choices.

(6.) The estimation that formed the basis for the residuals was performed slightly differently from that shown in the tables. Instead of using a multinomial estimation technique, the probability of an increase was estimated as a binomial logit, as was the probability of a decrease. This avoids the problem of ordering the realizations. The residual here is whether a discount rate recommendation change is in, minus the predicted probability of the change being in. Although the dependent variable is not continuous, measuring the residuals in this way is a crude approach to examining the correlation across the banks of any omitted variable.

(7.) Note that recommendations for the discount rate affect the actual discount rate only if the Board of Governors approves the recommendation. Similarly, votes by district presidents could be dissents rather than Votes with the majority, suggesting that monetary policy might not he affected by this possible relationship.

(8.) It is assumed here that the discount rate recommendation represents the desired monetary policy of whoever sets it. It is possible that the two different instruments have different purposes so that even if they were set by the same person or committee, the two decisions might depend on different variables. This possibility is discussed in detail later in the paper.

(9.) The regional employment growth is insignificant regardless of whether its national counterpart is included. Also, including the spread between the federal funds rate and the discount rate in all these equations had little effect on the estimation. The interpretation of the coefficients on this spread is not straightforward, so they are not presented in the estimation results.

(10.) The unemployment rate for the district, which is a weighted average of the unemployment rates of the states, is used in the analysis. Alternatively, the unemployment rate for the largest state or the city in which the district bank is located could have been used. However, the board of directors of each district hank is constructed to represent the different states in the district. Members are chosen to ensure that the variety of states are represented. Furthermore, since the major state in the district usually represents a very large share of the district, the district unemployment rate will be highly correlated with the state rate; in fact, many districts are dominated by one state. Finally, the economies in the different states in the region tend to move together. For these reasons, the indicators at the district level seem to be the most appropriate.

(11.) Joining the Federal Reserve Bank of St. Louis are the Cleveland and Richmond Feds. Monetarism at the Federal Reserve practically started at the St. Louis Fed. The St. Louis Model found in Anderson and Carlson (1970) is a simple monetarist model of the economy, and the president at the time was a strong advocate of the monetarist school (Francis 1970). The question of which other banks to label as monetarist is more debatable. Richmond was included because it offered a consistently monetarist voice throughout the 1970s and early 1980s (Humphrey 1974, 1975; Hetzel 1976; Cullison 1982); even the current president coauthored a sympathetic article (Broaddus and Goodfriend 1984). Cleveland's monetarist voice was less strong but was there (Carlson 1981). Altering the exact composition of this group has no substantive effect on the results.

(12.) Either the discount rate recommendation or the vote of the other bank presidents would capture the effect of the regional information on the other banks' voting. The discount rate is chosen because it is exogenous to the FOMC voting; it avoids any complications due to group dynamics. As a sidelight, including the other bank discount rate recommendations might also capture the effects of some omitted variables. The inclusion of the other banks' discount rate recommendations did not affect the significance of a bank's board's recommendation on its own president's vote. Note that an omitted variables explanation would explain episodes where there was consensus, not episodes where a bank listened to one bank but not another.

References

Anderson, Leooard C., and Keith M. Carison. 1970. A monetarist model for economic stabilization. Federal Reserve Bank of St. Louis Review (April):7-25.

Broaddus, Alfred, and Marvin Goodfriend. 1984. Base drift and the longer run growth of Ml: Experience from a decade of monetary targets. Federal Reserve Bank of Richmond Economic Review (November/December):3-14.

Carlson, Joho B. 1981. The monetary base, the economy, and monetary policy. Federal Reserve Bank of Cleveland Economic Review (spring):2-13.

Cullison, William E. 1982. Money, the monetary base, and nominal GNP. Federal Reserve Bank of Richmond Economic Review (May/June):3.-13.

Francis, Darryl R. 1970. The new, new economics and monetary policy. Federal Reserve Bank of St. Louis Review (January):5-9.

Hetzel, Robert. 1976. Predicting the rate of inflation in 1976. Federal Reserve Bank of Richmond Economic Review (January/February): 13-8.

Hoskins, Lee. 1991. Preface. Journal of Money, Credit, and Banking 23(3, Pt.2):iii.

Humphrey, Thomas M. 1974. The quantity theory of money: Its historical evolution and role in policy debate. Federal Reserve Bank of Richmond Economic Review (May/June):2-19.

Humphrey, Thomas M. 1975. A monetarist model of the inflationary process. Federal Reserve Bank of Richmond Economic Review (November/December): 13-23.

McNees, Stephen K. 1986. Modeling the Fed: A forward-looking monetary policy reaction function. Federal Reserve Bank of Boston, New England Economic Review (November/December):3-8.

MeNees, Stephen K. 1992. A forward-looking monetary policy reaction function: Continuity and change. Federal Reserve Bank of Boston, New England Economic Review (November/December):3-13.

McNees, Stephen K. 1993. The discount rate: The other tool of monetary policy. Federal Reserve Bank of Boston, New England Economic Review (July/August):3-22.

Peek, Joe, and Eric Rosengren. 1995. The capital crunch: Neither a borrower nor a lender be. Journal of Money, Credit, and Banking 27(3):625-38.

Rush, Mark, Gordon Sellon, and Li Zhu. 1994. The role of the discount rate in monetary policy. Federal Reserve Bank of Kansas City Working Paper 94-1.

Thornton, Daniel T. 1992. The market's reaction to discount rate changes: What's behind the announcement effect? Federal Reserve Bank of St. Louis Working Paper 92-003A.

Tootell, Geoffrey M. B. 1991. Regional economic conditions and the FOMC votes of district presidents. Federal Reserve Bank of Boston, New England Economic Review (September/Qctober):3-16.

Tootell, Geoffrey M. B. 1997. How farsighted is the FOMC? Federal Reserve Bank of Boston, New England Economic Review (January/February):49-65.

(a.)[Q.sup.E](1) and [P.sup.E](1) are the one-quarter-ahead Green Book forecasts of output growth and inflation. UR(-1) is the last known unemployment rate at the time of the meeting. M is the three-month average of lagged M1 growth. The numbers in parentheses are t-statistics.

(a.)[Q.sup.E](1), [P.sup.E](1), and U[R.sup.E](1) are one-quarter-ahead Green Book forecasts of output growth, inflation, and the unemployment rate. [Q.sup.E](1,2) and [P.sup.E](1,2) are the Green Book forecasts of output growth and inflation over the next six months. [Q.sup.E](3,4) and [P.sup.E](3,4) are the Green Book forecasts of output growth and inflation over the six months starting six months from now. M is the lagged three-month average of M1 growth. UR(- 1) is the lagged unemployment rate. CORE(- 1) refers to the lagged three-month change of the core CPI. The numhers in parentheses are t-statistics.

(a.)[Q.sup.E](1) and [P.sup.E](1) are the one-quarter-ahead Green Book forecasts of output growth and inflation. M is the lagged three-month average of Ml growth. UR(-l) is the lagged unemployment rate. DR = 1 if the policy is a discount rate recommendation and zero if the policy is the FOMC vote. The numbers in parentheses are t-statistics.

(a.)[Q.sup.E](1) and [P.sup.E](1) are the one-quarter-ahead Green Book forecasts of real output growth and inflation. M is the lagged three-month average of M1 growth. UR (-1) refers to the lagged unemployment rate. The numbers in parentheses are t-statistics.

(a.)For brevity, the coefficients for the other variables have not been inc1uded. The discount rate variables rise as the recommendation moves toward tighter policy. The numbers in parentheses are t-statistics.

In the United States, private citizens play little direct role in policymaking. The directors of the boards of the regional Federal Reserve banks are an apparent exception to this rule. These directors recommend changes in the discount rate, although the Board of Governors decides whether to act on the recommendations. These directors would have greater influence if they affected the FOMC votes of their district bank presidents. This paper shows that the FOMC votes of the regional bank presidents are strongly correlated with the discount rate recommendation of their bank's board. Several alternative explanations for the correlation are then examined.

1. Introduction

The directors who sit on the boards of the 12 district Federal Reserve banks are the rare private citizens who have a direct role in government decision making. These nine business, labor, financial, and community leaders are neither elected nor appointed by elected officials; as a result, their goals could differ from those of the nation as a whole. Although these directors do not actually set policy, they do recommend changes in the discount rate to the Board of Governors of the Federal Reserve System. Thus, on the surface their role in monetary policy deliberations is only advisory. However, their influence could in fact be substantially more important if they actually affected the Federal Open Market Committee (FOMC) votes of their respective Reserve bank presidents. This paper examines this more significant link to monetary policy by testing the relationship between the discount rate recommendation of the Reserve bank's board of directors and the vote of that district's bank president at the FOMC. It is shown that the FOMC votes of Reserve bank presidents are significantly correlated with their board's current discount rate recommendation.

Given the distinct responsibilities of the district board and the bank president, it is entirely possible that the president's vote would differ from the discount rate recommendation. By law, each Reserve bank's board of directors must make a discount rate recommendation to the Board of Governors every two weeks. On the other hand, every eight weeks a Reserve bank president must decide on his or her own view of the proper course for monetary policy at the FOMC meeting. De jure, there is no reason that the two decisions should conform. De facto, it is unclear whether, or by how much, the local board's discount rate recommendation influences the president's FOMC deliberations or whether or by how much the district president influences the local board's discount rate recommendation. This paper presents evidence that the two decisions are made separately but that the discount rate recommendation does help explain the president's FOMC vote.

For several reasons, one might expect the determinants of the local board's discount rate recommendation to differ from those of the president's FOMC vote. The local boards may have concerns and perspectives that deviate from those of FOMC members in general and their own president in particular. For example, members of the district board may disproportionately weight their region's economic performance, while the bank's president may have a stronger national focus. In fact, McNees (1993) shows that several district boards respond to regional economic conditions, while Tootell (1991) finds that district bank presidents do not base their FOMC votes on the economic performance of their regions. It is also possible that the district banks' boards of directors have different national goals than members of the FOMC; for example, local boards may disagree with their president on the relative costs of inflation and unemployment. Finally, the district board's forecast could systematically diverge from that of its pr esident. Any one of these possible causes could produce a schism between the discount rate recommendation of the bank's board of directors and the FOMC votes of its president.

On the other hand, there are several reasons why the two decisions should be highly correlated. As a national instrument, the FOMC vote of a bank president should depend on national economic conditions. The performance of the economy should be highly correlated with the variables, be they national, regional, or industry specific, that determine the local boards' discount rate recommendations. It is shown, however, that the macrovariables thought to be significant determinants of FOMC behavior do not explain all the correlation between a Reserve bank's discount rate recommendation and its president's FOMC vote. Alternatively, a bank's discount rate recommendation and its president's FOMC vote might be correlated because either the local board or the president is, essentially, making both decisions. However, evidence presented in this paper suggests that neither the district bank's board nor its president dominates both decisions. Finally, the district presidents and their boards may tend to agree on monetary policy and disagree with the rest of the Federal Reserve System because they share an outlook for the economy that differs from the rest of the FOMC, because they share goals at variance with the rest of the FOMC, or because they use a different model than the FOMC to analyze the data. All three of these explanations are explored in this paper.

The next section briefly describes the data. Section 3 then examines the correlation between the local board's discount rate recommendation and the bank president's FOMC vote. The discount rate recommendation is shown to add information to the explanation of the bank president's FOMC vote even when the relevant national and regional data are included. No obvious omitted macrovariable accounts for this result. Section 4 examines the possible sources of the discount rate recommendation's effect on the presidents' FOMC votes. The evidence suggests that the district bank's board of directors does influence, but does not control, the FOMC vote of the bank's president. Section 5 concludes.

2. The Data

The FOMC votes of district presidents and the discount rate recommendations of the regional boards are merged with data describing the current and expected future state of the economy. The FOMC votes were recovered from the minutes of the FOMC meetings, published approximately six weeks after each gathering. These minutes, and particularly the directive, outline the policy adopted and the vote of each FOMC member on that action. The wording of the directive has evolved over time; thus, at certain times it was easier to glean the course of monetary policy from the directive than at others. [1] Once the course of monetary policy is translated from the directive, judging the monetary policy action voted by each FOMC member is straightforward. When a member dissents from the chosen policy, the dissent is explained and the desired policy articulated. At any one meeting, seven governors and five presidents are permitted to vote, so in general 12 votes are registered. Finally, it should be emphasized that the direc tive provides only the direction of policy--either to tighten policy, loosen it, or keep it constant--and does not give the magnitude of any policy change. As a result, the dependent variable is discrete and trichotomous.

The data on the discount rate recommendation for each bank back to 1974 were retrieved from the discount rate minutes of the Federal Reserve. Every week the Board of Governors meets to consider action on the discount rate. If a recommendation is in, and it is in the direction that the Board of Governors desires, the members can vote to alter the rate. The minutes of that meeting provide information on any changes in the discount rate and on which banks recommended what change at the time of the meeting.

Several problems arise when merging data on each bank president's monetary policy vote with his or her board's discount rate recommendation. The district bank's discount rate recommendation is updated every two weeks. The FOMC, on the other hand, met roughly once a month prior to 1980 and has met eight times a year since. The higher frequency of the discount rate series must be altered to conform to the FOMC frequency. Since this paper examines the effect of a bank's discount rate recommendation on the bank president's FOMC vote, only the recommendation in at the time of the FOMC meeting is analyzed. Furthermore, discount rate recommendations made during periods when the district bank president was not a voting member of the FOMC cannot be used to examine the effect of these recommendations on the president's FOMC vote. Even though a discount rate recommendation is in, the district president's reaction to that recommendation can be quantified only when he or she votes at the FOMC.

The discount rate recommendations and the FOMC votes should, in general, be highly correlated since the Reserve banks' boards of directors and the members of the FOMC are reacting, certainly in part, to the same economy. Thus, when examining the effect of any variable on the monetary policy votes of FOMC members or on the local boards' discount rate recommendations, it is necessary to account for the state of the economy at the time these actions were taken. Since monetary policy works with long lags, changes in monetary policy should be determined by expectations about the future state of the economy. The expectations of the district presidents and their boards are instrumented for using the economic forecasts of the Federal Reserve Board staff, and these are circulated immediately prior to every FOMC meeting in the "Green Book." [2] Contained in the Green Book are Reserve Board staff forecasts of real gross national product (GNP) and its components as well as the unemployment rate and several measures of i nflation. Since the discount rate minutes are spotty before 1974 and the Green Book forecasts become available to the public only with a five-year lag, the sample covers the years 1974 to 1993.

McNees (1986, 1992) and Tootell (1991, 1997) show that the forecasted variables that affect monetary policy include the growth of real GNP, the unemployment rate, and the rate of inflation. Lagged values of the target variables are also examined to ensure that the results are not an artifact of imposing forward-looking behavior on all the policymakers. Including these macrovariables in the analysis ensures that any remaining correlation between the FOMC votes of the district presidents and the discount rate recommendations of their local boards is independent of the current and expected state of the economy.

3. Discount Rate Recommendations and FOMC Votes

All 12 Reserve banks are pooled in the analysis, although this constraint is relaxed later in the paper. The discount rate recommendation is also transformed into a trichotomous discrete variable equal to --1 if the local board advocates a reduction in the discount rate, zero if it advocates no change in the rate, and 1 if it advocates an increase in the rate. [3]

Because the dependent variable takes on only three discrete values, least squares estimation would produce inefficient estimates of the coefficients and inconsistent estimates of their standard errors. For this reason, the presidents' voting function is estimated as a multinomial logit. Although multinomial logit estimation of the coefficients and their standard errors produces efficient and consistent estimates, the coefficients are not as easily interpreted as those produced by least squares. The nonlinearity of the logit function results in the estimated effect of a variable on the FOMC members' voting probabilities depending on the values of the other right-hand-side variables. Thus, the economic significance of any variable must be measured at a given value of the other explanatory variables, usually selected to be their means. Finally, in order to identify the parameters in the model, the coefficients on the probability to keep policy unchanged are set equal to zero; as a result, the equations estimate the probability of voting to tighten or loosen policy relative to a vote of no change.

The coefficients from a multinomial logit estimation of the Reserve bank presidents' FOMC votes on a constant and their respective boards' discount rate recommendations are presented in the first column of Table 1. The top panel presents the coefficients measuring the effects of the explanatory variables on the probability of the presidents voting for a tightening in policy relative to the probability of voting for no change, while the bottom panel presents the coefficients estimating the effects of the independent variables on the probability of the presidents voting for a loosening in policy relative to the probability of voting for no change. In column 1, the coefficients on the discount rate recommendation reveal that presidents' votes at the FOMC are highly correlated with their bank's discount rate recommendation in at the time of the FOMC meeting. A bank president is much more likely to vote for a tightening and much less likely to vote for a loosening the more restrictive is the local board's discount rate recommendation. As mentioned earlier, this result should come as no surprise since the local boards of directors and the district presidents tend to react to the same economic variables; it is probable that similar expectations about the same macrotargets are driving both the discount rate recommendation and the FOMC vote.

To account for these expectations, column 2 of Table 1 assumes that the district presidents target some form of nominal GNP; policy depends on the expected growth in real GNP and the forecast of inflation. These forecast variables have the expected effect on the FOMC votes of the presidents. High expected inflation significantly increases the probability of tightening, and rapid real output growth significantly lowers the probability of loosening. Yet even after controlling for the expected state of the economy, the discount rate recommendation remains a significant determinant of the president's FOMC vote. If the local board's discount rate recommendation and the bank president's FOMC vote were merely reacting independently to the forecasted growth in prices and real output, then the discount rate recommendation should add no information to an equation explaining the FOMC votes that included the forecasts of these two variables. Inclusion of these forecasts, however, has almost no effect on the estimated pa rtial correlation between the discount rate recommendation and the district president's FOMC vote; the local board's discount rate recommendation is still statistically and economically significant in predicting the district president's vote at the FOMC.

The significance of the discount rate recommendation does not seem to arise from a simple misspecification of the presidents' voting function. In column 3, the coefficients from a less parsimonious model only reinforce the conclusion that the discount rate recommendation of each bank's board is significantly related to the FOMC vote of that bank's president. Money growth and the last known unemployment rate are added to the simple nominal GNP targeting rule. Including a monetary aggregate in the voting function makes sense if some district presidents are monetarists or if some district presidents believe that past money growth improves the Green Book's forecast of inflation. The unemployment rate should be included if it, rather than real GNP growth, is in the bank presidents' objective function. [4] The coefficients on the economic variables in column 3 are, for the most part, significant and correctly signed. In general, higher expected real growth, higher forecasted inflation, more rapid money growth, and lower expected unemployment increase the probability that the FOMC will vote to tighten policy and decrease the probability that it will vote to loosen policy. Yet the coefficients on the discount rate recommendation remain large and statistically significant. Adding these other variables to the FOMC voting function seems to have little effect on the size or significance of the coefficient on the discount rate recommendation.

The equations presented in Table 1 are clearly not exhaustive, and a misspecification could still be causing the significance of the discount rate recommendation. It is possible that the FOMC has a forecast horizon different from the one modeled in Table 1; the committee could be more forward or more backward looking. The significance of the discount rate recommendation in Table 1 may simply be due to a misspecification of the horizon to which the FOMC reacts. It is also still possible that some other macrovariable may be missing from the analysis. Both of these hypotheses will be examined in turn.

Different Forecast Horizon

The inclusion of only the one-quarter-ahead forecast may miss an important dynamic element of the economy that both the FOMC and the local board of directors are considering. Policymakers care about the economy's momentum and its performance over a longer time period than one quarter. One way to capture this longer-run performance is to include either lagged data or longer-horizon forecasts.

Table 2 explores the effects of altering the forecast horizon in the voting function. The specification in the first two equations is similar to that in column 3 of Table 1, except now explicit forward-looking data on the unemployment rate and backward-looking data on the core inflation rate are included. The first column of Table 2 provides some evidence that Reserve bank presidents are partially backward looking on core inflation. Lagged inflation has a significant and correctly signed effect on the probability of loosening but an incorrectly signed effect on the probability of tightening. However, the collinearity of the lagged and forecasted inflation measures makes it difficult to distinguish the effects of backward- and forward-looking inflation. [5]

To ensure that the significance of the discount rate recommendation in the equations in Table 1 is not accounted for by the omission of lagged data, Equation 2 adds the discount rate recommendation to an equation that includes the backward-looking variables. The coefficients on the discount rate recommendation of the bank's board are still large and statistically significant well beyond the 1% level. The omission of the lagged variables does not explain the significance of the discount rate recommendation.

Altering the specification to include forecasts as far as one year ahead also fails to diminish the importance of the discount rate recommendation on the presidents' FOMC vote. In column 3, the presidents' voting function is reestimated, including the forecasts of real GNP growth and inflation over the next six months and the six months after that, along with the backward-looking variables. As discussed in Tootell (1997), this specification smoothes out the quarterly noise in the forecast while examining the reaction to longer-term forecasts. The coefficient estimates presented in column 3 reveal that the longer-run forecasts are important, but the inclusion of all these variables has little effect on the strength of the relationship between the local board's discount rate recommendation and that bank president's FOMC vote. The importance of the discount rate recommendation to the president's FOMC vote does not appear to be an artifact of the forecast horizon examined in these voting functions.

Omitted Macrovariables

Although the forecasts should incorporate all the relevant macroinformation, and the ultimate goals of monetary policy seem to be relatively few and agreed on, some omitted macro-variable may still be causing the discount rate recommendations and the presidents' FOMC votes to move together. One approach to testing for omitted macrovariables is to examine the effects on Fed policy of a wide range of potential candidates, as is done in McNees (1986, 1992). Since the final specification in McNees (1992) looks very much like the specifications examined here, this approach will not be pursued. As an alternative, the coincidence of discount rate recommendations across districts is examined. If serious omitted macrovariables exist, agreement on their macroeffects should be widespread, and the discount rate recommendations should be highly correlated both across district banks and with actual FOMC policy changes.

In fact, most recommendations for discount rate changes were advocated without widespread agreement across the different Reserve bank boards of directors or across a majority of the FOMC. Over 50% of all recommended changes in the discount rate were not followed by an FOMC policy change in the recommended direction. When the discount rate recommendation did forecast FOMC policy changes, much of the coincidence could be accounted for by the current and expected state of the economy captured by the economic variables examined in the previous tables. Furthermore, when at least one district bank had a recommendation in to change the discount rate, the average number of banks with a recommendation in for change was only three. A majority of the time, the FOMC policy overall did not appear to agree with the recommended change in the discount rate, nor did the other banks' boards, suggesting that a pressing omitted macroevent probably does not explain the effect of the discount rate recommendation on the FOMC vote.

As mentioned previously, this low raw correlation of the recommendations for changes in the discount rate across district banks fails to take into account the state of the economy. If variables omitted from the estimation of the voting function were driving every Reserve bank's recommendation in a similar direction, then the correlation across banks of the residual from an estimation of the discount rate recommendation on the included macrovariables should be high. The average correlation across Reserve banks of this residual is 0.33, and the average multiple correlation of one bank's residuals with all banks is 0.42. [6] The relationship is positive, as would be expected if omitted macrovariables explained the discount rate's importance in the voting equation. However, the correlation is not very high. In fact, most of the times when more than one Reserve bank was in with a discount rate recommendation, the macroforecasts would have predicted such a recommendation; the variables in the estimation account fo r much of any correlation seen across the different Reserve banks' discount rate recommendations. Most of the movement of the discount rate recommendation unaccounted for by the economic data in these voting functions is not explained by the behavior of the other Reserve banks' residuals; it appears to be largely idiosyncratic, not shared.

And, what is also important, the district presidents responded to the idiosyncratic recommendations. The bank president's FOMC vote did not react to the discount rate recommendation only when the rest of the FOMC concurred with the recommendation. Bank presidents dissented about 20% of the time when their board had a recommendation in for a change and the FOMC did not change policy in that direction; this is a much higher frequency than the average dissent rate of 6%. In fact, almost 50% of all dissents by presidents occurred when their board had a discount rate request in for a policy change different from the policy action taken by the FOMC. The importance of the discount rate recommendation, even when the FOMC disagreed with it, suggests that the results are not driven by an omitted macrovariable.

The correlation of the local board's discount rate recommendation with its president's FOMC vote does not seem to be due to macrovariables excluded from the analysis. Not only should the forecasts capture most relevant macroevents, but neither the discount rate recommendations nor the unexplained part of the recommendations is highly correlated across districts. It appears that some shared perspective of the local board and the district president is affecting the FOMC votes of the bank presidents. The next section examines the possible nature of this shared perspective.

4. The Nature of the Idiosyncratic Information

Although the local board's discount rate recommendation helps predict its president's FOMC vote, the nature of the link between the two policy actions is not immediately clear. Certainly there is no necessary correspondence between the two, as the regional boards can recommend any discount rate they want and the district presidents can vote any way they wish at FOMC. However, several explanations for the correlation are possible. First, the local board's recommendation may directly affect the president's FOMC vote. [7] Alternatively, the reverse could be true; the president could be influencing the discount rate recommendation of the local board. In this case, the discount rate recommendation might reveal the president's intention before he or she gets to the FOMC. Finally, the correlation may be picking up a joint effect on the outlook of both the president and the local board, that may differ from the outlook of the rest of the FOMC; for example, a district president's forecast of, goals for, or model of t he economy may have common elements with those of his or her board of directors that are not shared by the rest of the FOMC. Any of these explanations could be responsible for the correlation between the discount rate recommendation and the district president's FOMC vote. Which of the above three explanations is more likely is examined here.

Are the Actions of the District Boards and Presidents Independent?

If the district board reacts differently to the economy when determining its recommendation for the discount rate than the district president does when formulating his or her FOMC vote, then estimates of the presidents' voting function should differ from estimates of the local board of directors' discount rate recommendation function. Conversely, if either the president or the local board decides both the discount rate recommendation and the FOMC vote, then the determinants of both decisions should be identical. [8]

To examine the similarity between these two decisions, the discount rate recommendation must be estimated as a function of the variables used to explain the presidents' FOMC votes. When the discount rate recommendation is a dependent variable, it is constructed to be comparable to that of the FOMC vote. Thus, in column 2 of Table 3, the top panel provides the estimated effects of the macrovariables on the local boards' probability of recommending an increase in the discount rate (a tightening) relative to a recommendation of no change, and the bottom panel gives the coefficients for the same variables' effects on the local board's probability of recommending a decrease in the discount rate (a loosening) relative to a recommendation of no change in that rate. The forecasts are used as instruments of the local board's expectations.

The first equation in Table 3 estimates the monetary policy stance of both the presidents and the local boards when the coefficients on the determinants of both the discount rate recommendation and the FOMC vote are constrained to be identical, which is the proper specification if the same person or committee made both decisions. The dependent variable in column 1 is the monetary policy vote of either the presidents at the FOMC or the desired discount rate move of the local board. Thus, the estimates presented in column 1 assume that the vote of a president for tighter or looser policy is determined by the same forces as the recommendation of the local board for a higher or lower discount rate. Columns 2 and 3 present the coefficients from the unconstrained estimation, where the coefficients on the presidents' FOMC votes are allowed to differ from those of the local boards' discount rate recommendation. The significance levels of the coefficients in columns 2 and 3 reveal that the two decisions depend on man y of the same variables. Yet it can be rejected well beyond the 1% level that the two sets of coefficients are the same. The results suggest that the two decisions are made differently, supporting the hypothesis that neither the bank president nor the district board of directors dominates both decisions.

However, it is possible that the discount rate recommendation and the federal funds rate serve different functions. Differences in the functions of the two instruments could explain differences in their estimated determinants. A fairly extensive literature explores the different roles that the discount rate and the federal funds rate may play. For example, Thornton (1992) and Rush, Sellon, and Zhu (1994) have examined the announcement value of discount rate changes, suggesting that the discount rate is used to signal a more permanent change in interest rates, while the federal funds rate is used to probe changes.

In fact, the evidence provides little support for the conclusion that the discount rate recommendation is used as an announcement device. A greater reluctance to use the discount rate should result in a different constant term in the discount rate recommendation equation compared to the FOMC voting function, but it should not produce different coefficients on the economic variables in these two functions. In fact, when the constant term is allowed to differ across the two decisions, the coefficients on the economic variables still differ significantly across the two decisions. On a more intuitive level, if the discount rate recommendation was meant to signal a long-run change in rates and the same person was making both decisions, then we should frequently see bank presidents voting to change the federal funds rate and not the discount rate but rarely see the opposite. In fact, 50% of the time that a recommendation for a change in the discount rate was in, it was not ratified by a change in FOMC policy, nor did the district president oppose the inaction--the opposite of the direction to be expected according to the announcement effect.

Alternatively, it is possible tat the discount rate recommendation represents the bank president's inclination for monetary policy, even though he or she may be unwilling to dissent at the FOMC. If the district president were reluctant to dissent but wanted to reveal policy inclinations in a more subtle way, one would expect to see many more recommendations for discount rate changes than FOMC votes for change. Conversely, one would not expect to see FOMC votes to change policy without a discount rate recommendation in. Yet over 60% of FOMC dissents occurred without a similar discount rate recommendation in from that president's local board. In fact, the district bank frequently had a recommendation in for a policy move in one direction while its president voted for a move in the opposite direction. The data are not consistent with the hypothesis that the president or the local board of directors made both decisions.

Finally, it must be emphasized that the previous literature examines the actions by the Board of Governors in setting the discount rate, not the desires of the regional boards of directors in recommending changes in this rate. Recommendations about the discount rate are not the same as changes in the discount rate. Even if the discount and federal funds rates serve different functions, it is not clear that the district boards of directors feel constrained by these differences. Local boards are likely to consider their recommendations as a signal for their desired federal funds rate changes. Since the Board of Governors actually controls the discount rate, the constraints caused by any nuances of the potentially different functions that the two instruments may serve are left for the Board of Governors to sort out. Since the funds rate, not the discount rate, directly affects the economy, it makes sense that the discount rate recommendation is used by the regional board of directors as a signal of its desired funds rate changes.

In an attempt to uncover the source of the different policy reactions between the presidents and their local boards, column 4 of Table 3 presents coefficient estimates of variables that interact a dummy variable indicating whether the monetary policy decision being made is a discount rate recommendation with all the other explanatory variables. Note that this sample is the same as that used in column 1 since the null hypothesis is that the two sets of coefficients are identical. In general, it appears that the monetary policy desired by the local boards of directors may be more activist than the policy voted on by the district bank presidents. The local boards seem to react much more strongly to the real economy, as the coefficients on the lagged unemployment rate and the expected growth in real output are larger in absolute terms. Some evidence also suggests that the local boards react more strongly to inflation. Again, if the president or the local boards simply controlled these two different instruments, it is not clear why these particular variables should explain their different uses. The results in Table 3 demonstrate that presidents at the FOMC respond differently to the state of the economy than do their local boards, suggesting that the discount rate recommendation is not simply foreshadowing the president's desires, nor is the president simply following the local board's orders.

Regional Information

Although the two decisions appear to be somewhat independent, the marginal predictive power of the discount rate recommendation on the president's FOMC vote suggests significant interrelation. The policy desires of the president and the local board might coincide while differing with the FOMC as a whole for several reasons. The most obvious explanation is that they both give more weight to their district's economic performance than its share in GDP. Tootell (1991) examined the role played by regional economic indicators in the FOMC voting of district presidents and found that regional economic conditions, as measured by employment growth and the unemployment rate in the district, added no marginal explanatory power to the national data. McNees (1993), however, found that regional economic indicators did affect the discount rate recommendation of some local boards. Some evidence suggests that region-specific economic information affects the votes of some local boards, but there is little evidence that these e ffects spill over into the FOMC votes of the presidents.

Table 4 merges these two findings by examining the effect of regional employment growth and unemployment rates on the formation of monetary policy. Column 1 presents the estimates of the effect of the Reserve district's employment growth on the local board's discount rate recommendation. Regional employment growth is not significant. [9] The second column shows the results when the district's unemployment rate, rather than its employment growth rate, is used as a proxy for the region's economic health. A higher regional unemployment rate, holding the national rate constant, does increase the probability that the local boards of directors recommend a lowering of the discount rate. The coefficient on the regional unemployment rate provides some evidence that local boards determine their discount rate recommendations on the basis of how well the regional economy is doing.

The results from an examination of the effect of the local economy on the district president's FOMC vote are presented in the next two columns. Neither the regional employment growth rate nor the unemployment rate is a significant determinant of how the district presidents will vote at the FOMC; the evidence suggests that the district presidents pay attention to the macrodata rather than the regional data. These results are robust to different specifications and consistent with the findings in Tootell (1991) and McNees (1993). [10] More important for this study, column 5 of Table 4 shows that the partial correlation of the discount rate recommendation with the FOMC vote of that district bank's president is unaffected by the inclusion in the analysis of the region's economic performance. The correlation between the discount rate recommendation and the president's FOMC vote is not due to shared concerns about the economic performance of the region.

The difference in the use of regional information across the two policy decisions also supports the conclusion that the two decisions are made by different groups. If the two decisions were made by the same people, there could be some reason to suspect that the constant term in the discount rate recommendation equation would differ from that in an FOMC voting equation; there would be little reason to expect the coefficients on the macroforecast variables to differ and no reason to expect that regional information should be important for the discount rate but not for the federal funds rate. While regional information does not explain the correlation between the discount rate recommendation and the presidents' FOMC vote, it does emphasize the difference in the objective function of the two groups making the decisions.

Models, Goals, and a Regional Outlook

The two decisions appear to be fairly independent, and although the correlation is not explained by region-specific economic conditions, regional information in a broader sense may be its origin. One possibility is that individual Reserve banks may have models of the economy or goals for monetary policy that differ from the rest of the FOMC. For example, the Federal Reserve bank of St. Louis has historically had a monetarist outlook, while the Federal Reserve bank of Cleveland tends to have a strong distaste for inflation (Hoskins 1991). Since the local board of directors plays an important role in the selection of the Reserve bank president, agreement between the local board and the president about the economic model and the appropriate goals for monetary policy may explain why their monetary policy reactions are partially correlated yet can be somewhat different from those of the other FOMC members.

To capture the effect of potential differences in goals, individual Reserve bank dummy variables were included in the estimation. Since the constant term is a linear combination of the coefficient estimates and these goals, estimating the logits with different regional bank constant terms permits the goals to differ between banks. Allowing the constants to vary across the district banks has no effect on the results; the discount rate recommendation of the local board continues to have a strong effect on the FOMC vote of the Reserve bank's president. To capture the possible effects of variation across banks in the model used, the most distinctive model difference is also examined--whether the bank is monetarist. The coefficient on money is allowed to differ for the three banks with the most consistent monetarist outlook. [11] Allowing this coefficient to vary fails to account for the size or significance of the coefficient on the discount rate recommendation. Differences in the banks' goals or models do not s eem to be the answer.

It is also possible that the outlook for the national economy could depend on the region. The national forecast could vary because of region-specific experience not associated directly with its economic performance. In this case, the Green Book forecasts would be missing the shadings of the outlook that depend on the region in which the policymaker lives. The credit crunch in New England in the early 1990s, as examined by Peek and Rosengren (1995), is one possible example of how the national outlook can depend on the region. In the early 1990s, a credit squeeze in New England had the potential to expand to the rest of the country since any district where real estate values were weak was subject to the shock. It is possible that the assessment of the probabilities that this effect would spread, and the assessment of the consequences if it did, may have been more dire in New England. Thus, at that time, both the local board and the bank president at the FOMC may have been more negative about the national econo my than were the Green Book forecasts and the rest of the FOMC.

As seen in the previous section, distinguishing the exact source of the differences in the shadings of the outlook is difficult and beyond the scope of this paper. However, one way to attempt to capture the regional outlook independent of the region's economic performance is to examine whether members of the FOMC find another region's or another bank's shading on the forecast compelling. Sticking to the credit crunch example, district presidents from areas where real estate values were beginning to fall may have found reservations about the economy based on the possible effect of the credit constraints more sympathetic. In this way, idiosyncratic differences about the national outlook could influence other FOMC members. This influence is measured in Table 5. The effect of each bank's discount rate recommendation on the FOMC vote of the other FOMC members, when the state of the economy and the discount rate recommendation of that bank are also taken into account, is shown. For brevity, Table 5 presents only t he coefficients measuring the effect of each bank's discount rate recommendation on all the other FOMC members.

The recommendations of local boards often do affect the other members of the FOMC, as would be true if the banks brought information to the meeting that the other FOMC members found persuasive. Ten of 24 such coefficients are correctly signed and statistically significant beyond the 10% level. As an example, note that the discount rate recommendation from New York significantly affects the probability of both tightening and loosening by the other FOMC members. New York's effect could be due to capital market information that many at the FOMC found important to the shading of the national outlook. The different outlooks or information could be the source of the partial correlation between the board's discount rate recommendation and the president's FOMC vote--their persuasiveness at the FOMC suggests the potential value of these different outlooks to the formation of monetary policy. [12]

5. Conclusion

A district's discount rate recommendation is clearly a powerful signal of that Reserve bank president's FOMC vote. Although the discount rate recommendation does depend on many of the same economic variables as the FOMC vote, it adds marginal predictive power for the FOMC vote beyond these macrovariables. Yet the discount rate recommendation is not simply the district president telegraphing his or her punch, and the FOMC vote is not the president simply following orders from his or her local board since the determinants of the two decisions differ significantly. The correlation seems to derive from idiosyncratic information not related to the macroforecasts in the Green Book but shared at the regional level. Different information appears to be emphasized in the different regions, and some evidence suggests that the FOMC as a whole finds this information useful.

The implications for monetary policy are important. If monetary policy were being disproportionately influenced by concerns about the regional economic performance or by narrow private interests, the federal structure could be suboptimal. There is little evidence that local conditions in these regions overly affect presidents' votes at the FOMC meetings. However, the meeting of the local boards and district presidents does seem to affect the presidents' FOMC votes, and this could be beneficial if it results in the presentation of different information, different points of view, or different forecasts at the FOMC meeting. In fact, there is evidence that the discount rate recommendations related to idiosyncratic information at the district level often affect the votes of the other FOMC members; this highlights one important benefit of the Federal Reserve System--its inclusion of different views for a full debate. In this way, the presentation of different perspectives at the FOMC justifies its structure as a c ommittee including representation from different regions.

(*.) Research Department, Federal Reserve Bank of Boston, 600 Atlantic Avenue, Boston, MA 02106-2076, USA; E-mail geoff.tootell@bos.frb.org.

The author wishes to thank Lynn Browne, Richard Kopcke, Stephen McNees, and two anonymous referees for their helpful comments. Valuable research assistance was provided by Rokeya Kahn and Peter Morrow, The analysis and conclusions of this paper are not necessarily endorsed by the Federal Reserve System or the Federal Reserve Bank of Boston.

Received December 1997; accepted August 1999.

(1.) Since 1997, the FOMC has released the explicit target for the federal funds rate after each meeting. Earlier, the directive called for maintaining, tightening, or easing either "reserve conditions" or the federal funds rate. The assessment of monetary policy from the directive depended on which of these alternatives was called for by the directive. Only during the monetary experiment of the early 1980s did judgment play an Important role in discerning the course of monetary policy.

(2.) The local boards of directors do not have access to the Green Book forecasts. Nonetheless, the Green Book projections are used as instruments for more widely available market forecasts. The results are similar when publicly available forecasts, suhs as those from DRI, are used. Any potential informational advantage contained in the Green Book would bias the results toward rejecting that the local board's discount rate recommendation had a partial correlation with the president's FOMC vote.

(3.) The discount rate recommendation is transformed this way for several reasons. First, the recommendations tend to be fairly discontinuous; they take on values in 25-basis-point multiples from 100 to --100. Generally, the recommendations are for 50-basis-point movements, so constraining the discount rate recommendation to take on one of only three values is not that restrictive. Also, it will be useful later to have the discount rate recommendation and the FOMC vote in comparable form. Finally, when separate dummies representing times when a Reserve bank had a recommendation in for a discount rate tightening and times when it had a recommendation in for a discount rate loosening were included in the analysis, the results were identical, and relaxing the assumption of a symmetric response did not increase the explanatory power of the estimation. As defined, as the variable increases, the probability of the president voting for a tightening should rise, and his or her probability of voting for a loosening sh ould fall.

(4.) If Okun's law holds with regularity, giving today's unemployment rate and the forecasted growth in gross domestic product (GDP) is equivalent to giving the forecast of the unemployment rate.

(5.)The pattern of the coefficients on inflation for the probability of loosening suggests that changes in inflation may be Important; however, there is little evidence of such a relationship for the probability of tightening. The estimation in the second and third columns of Table 2 allows the specification to differ for the two different choices.

(6.) The estimation that formed the basis for the residuals was performed slightly differently from that shown in the tables. Instead of using a multinomial estimation technique, the probability of an increase was estimated as a binomial logit, as was the probability of a decrease. This avoids the problem of ordering the realizations. The residual here is whether a discount rate recommendation change is in, minus the predicted probability of the change being in. Although the dependent variable is not continuous, measuring the residuals in this way is a crude approach to examining the correlation across the banks of any omitted variable.

(7.) Note that recommendations for the discount rate affect the actual discount rate only if the Board of Governors approves the recommendation. Similarly, votes by district presidents could be dissents rather than Votes with the majority, suggesting that monetary policy might not he affected by this possible relationship.

(8.) It is assumed here that the discount rate recommendation represents the desired monetary policy of whoever sets it. It is possible that the two different instruments have different purposes so that even if they were set by the same person or committee, the two decisions might depend on different variables. This possibility is discussed in detail later in the paper.

(9.) The regional employment growth is insignificant regardless of whether its national counterpart is included. Also, including the spread between the federal funds rate and the discount rate in all these equations had little effect on the estimation. The interpretation of the coefficients on this spread is not straightforward, so they are not presented in the estimation results.

(10.) The unemployment rate for the district, which is a weighted average of the unemployment rates of the states, is used in the analysis. Alternatively, the unemployment rate for the largest state or the city in which the district bank is located could have been used. However, the board of directors of each district hank is constructed to represent the different states in the district. Members are chosen to ensure that the variety of states are represented. Furthermore, since the major state in the district usually represents a very large share of the district, the district unemployment rate will be highly correlated with the state rate; in fact, many districts are dominated by one state. Finally, the economies in the different states in the region tend to move together. For these reasons, the indicators at the district level seem to be the most appropriate.

(11.) Joining the Federal Reserve Bank of St. Louis are the Cleveland and Richmond Feds. Monetarism at the Federal Reserve practically started at the St. Louis Fed. The St. Louis Model found in Anderson and Carlson (1970) is a simple monetarist model of the economy, and the president at the time was a strong advocate of the monetarist school (Francis 1970). The question of which other banks to label as monetarist is more debatable. Richmond was included because it offered a consistently monetarist voice throughout the 1970s and early 1980s (Humphrey 1974, 1975; Hetzel 1976; Cullison 1982); even the current president coauthored a sympathetic article (Broaddus and Goodfriend 1984). Cleveland's monetarist voice was less strong but was there (Carlson 1981). Altering the exact composition of this group has no substantive effect on the results.

(12.) Either the discount rate recommendation or the vote of the other bank presidents would capture the effect of the regional information on the other banks' voting. The discount rate is chosen because it is exogenous to the FOMC voting; it avoids any complications due to group dynamics. As a sidelight, including the other bank discount rate recommendations might also capture the effects of some omitted variables. The inclusion of the other banks' discount rate recommendations did not affect the significance of a bank's board's recommendation on its own president's vote. Note that an omitted variables explanation would explain episodes where there was consensus, not episodes where a bank listened to one bank but not another.

References

Anderson, Leooard C., and Keith M. Carison. 1970. A monetarist model for economic stabilization. Federal Reserve Bank of St. Louis Review (April):7-25.

Broaddus, Alfred, and Marvin Goodfriend. 1984. Base drift and the longer run growth of Ml: Experience from a decade of monetary targets. Federal Reserve Bank of Richmond Economic Review (November/December):3-14.

Carlson, Joho B. 1981. The monetary base, the economy, and monetary policy. Federal Reserve Bank of Cleveland Economic Review (spring):2-13.

Cullison, William E. 1982. Money, the monetary base, and nominal GNP. Federal Reserve Bank of Richmond Economic Review (May/June):3.-13.

Francis, Darryl R. 1970. The new, new economics and monetary policy. Federal Reserve Bank of St. Louis Review (January):5-9.

Hetzel, Robert. 1976. Predicting the rate of inflation in 1976. Federal Reserve Bank of Richmond Economic Review (January/February): 13-8.

Hoskins, Lee. 1991. Preface. Journal of Money, Credit, and Banking 23(3, Pt.2):iii.

Humphrey, Thomas M. 1974. The quantity theory of money: Its historical evolution and role in policy debate. Federal Reserve Bank of Richmond Economic Review (May/June):2-19.

Humphrey, Thomas M. 1975. A monetarist model of the inflationary process. Federal Reserve Bank of Richmond Economic Review (November/December): 13-23.

McNees, Stephen K. 1986. Modeling the Fed: A forward-looking monetary policy reaction function. Federal Reserve Bank of Boston, New England Economic Review (November/December):3-8.

MeNees, Stephen K. 1992. A forward-looking monetary policy reaction function: Continuity and change. Federal Reserve Bank of Boston, New England Economic Review (November/December):3-13.

McNees, Stephen K. 1993. The discount rate: The other tool of monetary policy. Federal Reserve Bank of Boston, New England Economic Review (July/August):3-22.

Peek, Joe, and Eric Rosengren. 1995. The capital crunch: Neither a borrower nor a lender be. Journal of Money, Credit, and Banking 27(3):625-38.

Rush, Mark, Gordon Sellon, and Li Zhu. 1994. The role of the discount rate in monetary policy. Federal Reserve Bank of Kansas City Working Paper 94-1.

Thornton, Daniel T. 1992. The market's reaction to discount rate changes: What's behind the announcement effect? Federal Reserve Bank of St. Louis Working Paper 92-003A.

Tootell, Geoffrey M. B. 1991. Regional economic conditions and the FOMC votes of district presidents. Federal Reserve Bank of Boston, New England Economic Review (September/Qctober):3-16.

Tootell, Geoffrey M. B. 1997. How farsighted is the FOMC? Federal Reserve Bank of Boston, New England Economic Review (January/February):49-65.

Discount Rate Recommendations and The FOMC Voting of District Presidents (1) (2) (3) Tightening [a] C -1.27 -2.33 -1.77 (-13.98) (-8.07) (-2.80) [Q.sup.E](1) 0.02 0.04 (0.89) (1.26) [P.sup.E](1) 0.17 0.17 (4.21) (4.33) UR(-1) -0.15 (-1.87) M 0.05 1.50 (2.85) Discount rate 1.66 (6.56) 1.40 recommendation (7.38) (5.92) Loosening [a] C -1.59 -1.55 -0.76 (-15.42) (-4.75) (-1.20) [Q.sup.E](1) -0.13 -0.13 -4.19) (-4.01) [P.sup.E](1) 0.03 -0.03 (0.71) (-0.53) UR(-1) 0.04 (0.44) M -0.10 (-4.43) Discount rate -1.25 -1.30 -1.30 recommendation (-5.20) (-5.14) (-4.98) Log of -776.2 -754.3 -736.0 likelihood Observations 923 923 923

(a.)[Q.sup.E](1) and [P.sup.E](1) are the one-quarter-ahead Green Book forecasts of output growth and inflation. UR(-1) is the last known unemployment rate at the time of the meeting. M is the three-month average of lagged M1 growth. The numbers in parentheses are t-statistics.

FOMC Voting: Effect of Forecast Horizon Dependent Variable Presidents' FOMC Votes Tightening [a] C -0.93 -1.45 -2.42 (-1.50) (-2.28) (-3.37) [Q.sup.E](1) 0.05 0.03 (1.94 (1.23) [Q.sup.E](1,2) -0.10 (-1.74) [Q.sup.E](3,4) 0.39 (4.66) [P.sup.E](1) 0.31 0.31 (4.86) (4.69) [P.sup.E](1,2) 0.13 (0.92) [P.sup.E](3,4) 0.34 (2.23) M 0.06 0.05 0.07 (3.17) (2.66) (3.23) UR(-1) 0.63 0.72 1.16 (1.24) (1.38) (2.08) [UR.sup.E](1) -0.92 -0.91 -1.51 (-1.83) (-1.76) (-2.72) CORE(-1) -0.10 -0.13 -0.08 (-2.15) (-2.73) (-1.36) Discount rate 1.47 1.48 recommendations (6.10) (5.86) Loosening [a] C -0.58 -0.56 1.04 (-0.93) (-0.89) (1.35) [Q.sup.E](1) -0.12 -0.11 (-3.58) (-3.47) [Q.sup.E](1,2) -0.35 (-5.16) [Q.sup.E](3,4) 0.12 (1.26) [P.sup.E](1) 0.10 0.13 (1.27) (1.54) [P.sup.E](1,2) 0.15 (0.85) [P.sup.E](3,4) -0.42 (-2.38) M -0.11 -0.11 -0.12 (-4.84) (-4.68) (-4.62) [UR.sup.E](1) -0.24 -0.53 -0.58 (-0.46) (-0.99) (-1.05) UR(-1) 0.28 0.54 0.58 (0.54) (0.99) (1.04) Loosening [a] CORE(-1) -0.13 -0.13 -0.03 (-2.35) (-2.37) (-0.40) Discount rate -1.31 -1.15 recommendation (-4.98) (-4.21) Log of -767.4 -728.9 -687.8 likelihood Observations 923 923 923

(a.)[Q.sup.E](1), [P.sup.E](1), and U[R.sup.E](1) are one-quarter-ahead Green Book forecasts of output growth, inflation, and the unemployment rate. [Q.sup.E](1,2) and [P.sup.E](1,2) are the Green Book forecasts of output growth and inflation over the next six months. [Q.sup.E](3,4) and [P.sup.E](3,4) are the Green Book forecasts of output growth and inflation over the six months starting six months from now. M is the lagged three-month average of M1 growth. UR(- 1) is the lagged unemployment rate. CORE(- 1) refers to the lagged three-month change of the core CPI. The numhers in parentheses are t-statistics.

Similarity of the Discount Rate Recommendation and the FOMC Vote Monetary Boards' Monetary Policy Policy Discount Rate Presidents' Vote and Dependent Variable Vote Recommendation FOMC Votes Interaction Effects Tightening [a] (1) (2) (3) (4) C -0.75 1.24 -1.19 -0.57 (-1.49) (1.26) (-1.94) (-1.11) [Q.sup.E](1) 0.09 0.17 0.05 0.05 (4.07) (4.04) (2.00) (1.97) [P.sup.E] (1) 0.20 0.23 0.21 0.19 (6.93) (4.43) (5.48) (5.24) M 0.06 0.07 0.06 0.06 (4.29) (2.57) (3.32) (3.27) UR(-1) -0.42 -0.90 -0.26 -0.34 (-5.91) (-5.74) (-3.28) (-4.54) [Q.sup.E](1)*DR 0.10 (2.13) [P.sup.E](l)*DR 0.08 (1.45) M*DR 0.006 (01.17) UR(-1)*DR -0.32 (-3.79) Loosening [a] C -1.50 -2.61 -0.75 -1.44 (-3.16) (-3.36) (-1.21) (-2.98) [Q.sup.E](1) -0.10 -0.06 -0.13 -0.12 (-4.30) (-1.51) (-4.22) (-4.06) [P.sup.E](1) -0.09 -0.17 -0.05 -0.02 (-2.36) (-2.43) (-0.98) (-0.41) M -0.08 -0.05 -0.10 -0.10 (-4.82) (-1.85) (-4.64) (-4.56) UR(-1) 0.13 0.24 0.08 0.14 (2.25) (2.50) (0.98) (2.14) [Q.sup.E](1)*DR 0.05 (1.08) [P.sup.E](1)*DR -0.20 (-2.71) M*DR 0.05 (1.43) UR(-1)*DR -0.02 (-0.19) Log of -1321.5 -484.8 -773.1 -1262.4 likelihood Observations 1846 923 923 1846

(a.)[Q.sup.E](1) and [P.sup.E](1) are the one-quarter-ahead Green Book forecasts of output growth and inflation. M is the lagged three-month average of Ml growth. UR(-l) is the lagged unemployment rate. DR = 1 if the policy is a discount rate recommendation and zero if the policy is the FOMC vote. The numbers in parentheses are t-statistics.

Regional Information and Monetary Policy Discount Rate Discount Rate FOMC Dependent Variable Recommendation Recommendation Vote Tightening [a] C 1.12 0.68 -3.71 (0.94) (0.58) (-4.53) [Q.sup.E](1) 0.17 0.16 -0.01 (3.60) (3.06) (-0.42) [P.sup.E](1) 0.23 0.29 0.21 (4.37) (5.25) (5.31) M 0.07 0.05 0.06 (2.49) (1.53) (3.16) UR(-1) -0.89 -0.95 0.02 (-4.91) (-4.12) (0.18) Regional employment growth -0.006 0.06 (-0.08) (1.05) National employment growth 0.02 0.28 (0.20) (3.21) Regional unemployment rate 0.13 (1.12) Discount rate recommendation Loosening [a] C -1.59 -1.57 -0.60 (-1.77) (-1.67) (-0.82) [Q.sup.E](1) -0.03 -0.24 -0.12 (-0.71) (-3.70) (-3.65) [P.sup.E](1) -0.17 -0.33 -0.04 (-2.45) (-3.03) (-0.85) M -0.04 -0.02 -0.10 (-1.68) (-0.70) (-4.47) UR(-1) 0.12 -0.17 0.06 (1.05) (-0.94) (0.59) Regional employment growth 0.001 0.08 (0.02) (1.32) National employment growth -0.18 -0.15 (-1.57) (-1.52) Regional unemployment rate 0.35 (2.69) Discount rate recommendation Log of likelihood -482.5 -344.0 -756.2 Observations 923 664 923 FOMC FOMC Dependent Variable Vote Vote Tightening [a] C -2.25 -2.72 (-3.32) (-3.89) [Q.sup.E](1) 0.12 0.09 (3.05) (2.34) [P.sup.E](1) 0.25 0.21 (5.94) (4.64) M 0.04 0.04 (2.20) (2.03) UR(-1) -0.12 -0.02 (-1.02) (-0.19) Regional employment growth National employment growth Regional unemployment rate -0.02 -0.02 (-0.26) (-0.25) Discount rate recommendation 1.32 (4.90) Loosening [a] C 1.56 1.38 (1.86) (1.66) [Q.sup.E](1) -0.32 -0.29 (-5.60) (-4.99) [P.sup.E](1) -0.33 -0.29 (-3.91) (-3.39) M -0.07 -0.07 (-2.88) (-2.83) UR(-1) -0.04 -0.02 (-0.23) (-0.13) Regional employment growth National employment growth Regional unemployment rate -0.03 -0.07 (-0.32) (-0.64) Discount rate recommendation -0.97 (-2.96) Log of likelihood -526.9 507.6 Observations 664 664

(a.)[Q.sup.E](1) and [P.sup.E](1) are the one-quarter-ahead Green Book forecasts of real output growth and inflation. M is the lagged three-month average of M1 growth. UR (-1) refers to the lagged unemployment rate. The numbers in parentheses are t-statistics.

Effect of Discount Rate Recommendations on Other FOMC Members Bank Tightening [a] Loosening [a] Boston 0.67 -0.19 (2.38) (-0.59) New York 1.12 -0.64 (4.33) (-1.99) Philadelphia -0.58 0.23 (-1.99) (0.62) Cleveland -0.52 -1.79 (-2.46) (-5.78) Richmond -0.83 -0.40 (-3.07) (-1.16) Atlanta 0.76 0.78 (3.61) (2.48) Chicago -0.41 0.22 (-1.86) (0.95) St. Louis 0.85 0.38 (3.67) (1.06) Minneapolis 0.69 -2.10 (2.56) (-7.31) Kansas City -0.62 0.68 (-2.19) (2.43) Dallas 0.06 -0.39 (0.30) (-1.68) San Francisco 0.26 -0.74 (1.28) (-3.47) Log of likelihood -1278.4 Observations 1761

(a.)For brevity, the coefficients for the other variables have not been inc1uded. The discount rate variables rise as the recommendation moves toward tighter policy. The numbers in parentheses are t-statistics.

Printer friendly Cite/link Email Feedback | |

Author: | B. Tootell, Geoffrey M. |
---|---|

Publication: | Southern Economic Journal |

Geographic Code: | 1USA |

Date: | Apr 1, 2000 |

Words: | 10219 |

Previous Article: | The Economics of Animal Farm. |

Next Article: | Welfare Policy: Cash Versus Kind, Self-Selection and Notches. |