Printer Friendly

The transmission channels of monetary policy: how have they changed?

Over the past two decades, important structural changes in the economy-resulting from institutional, regulatory, and technological developments--may have altered the nature and stability of the channels through which monetary policy affects the level of economic activity in the short run. For example, the abandonment of fixed exchange rates in the early 1970s and the integration of world capital markets have increased the scope for monetary policy to affect a growing traded-goods sector. The introduction of adjustable-rate mortgages, the removal of regulatory ceilings on deposit rates, and the development of secondary mortgage markets may have altered the interest sensitivity of residential construction activity. The runup in corporate debt--a consequence of the surge of takeovers and leveraged buyouts in the 1980s--may have altered the response of business capital spending plans to interest rates. And the increased sensitivity to interest rates of household interest income, owing to the removal of deposit rate ceilings, and of household interest payments, owing to the growing share of adjustable-rate financial liabilities, may have had some bearing on consumption behavior. Such developments could greatly affect not only the ways in which monetary policy influences the economy but also the strength of its overall influence.

Assessing the scope and magnitude of possible structural changes in the economy that may have affected the transmission channels of monetary policy is a difficult task. (1) The approach employed in this article is to use the MPS model of the U.S. economy to examine whether the key links between monetary policy and economic activity appear to have changed appreciably over the past decade. The MPS model is a large-scale econometric model of the United States that reflects mainstream macroeconomic theory and standard econometric practice. It was developed in the late 1960s in collaboration with university economists and has been maintained and updated by the staff of the Federal Reserve Board over the past twenty years. (2)

The approach taken here is not without its shortcomings. Evidence that the equations in the model have not remained constant over time could reflect either structural changes in the economy or some underlying misspecification of the relationships in the equations. To make the case for structural change rather than misspecification, we augment the results of standard statistical tests with evidence that the properties of the equation are consistent with accepted economic theory and that observed shifts are consistent with the specific hypotheses regarding change. In addition, the simulation and forecasting properties of this model have been well documented and thus provide some perspective on the strengths and weaknesses of the model, which aids in interpreting the tests undertaken for structural change.

The first part of this article describes the three main channels through which monetary policy actions affect real spending in the model--the cost of capital, the value of assets, and the foreign exchange value of the dollar. Model simulations quantify the relative importance of each channel in the current structure of the model, which is estimated using economic data from the past thirty years.

The second and third sections examine, in turn, the effect of monetary policy on financial market variables and the influence of those variables on spending decisions. Specifically, the second section examines whether the relationships underpinning the demand for money, the' term structure of interest rates, the value of corporate stock, and the foreign exchange value of the dollar changed appreciably over the 1980s. The third section explores whether the manner in which these financial variables affect consumption and investment has changed in the past ten years. To study the question of structural change, these sections present results of statistical tests aimed at identifying changes in the relationship between economic variables. (For an alternative approach to the issue of identifying structural change, see the appendix.)

In the final section, simulations of two versions of the model are used to examine whether the changes of the past decade have, on net, raised or lowered the sensitivity of output to changes in monetary policy. The two versions represent the distinct responses of financial markets and real spending that have been found to be significantly different over the pre- and post-1980 periods.

The main findings of the article are as follows: The sensitivity of aggregate output to changes in monetary policy is about the same now as it was in the 1960s and 1970s, until about the third or fourth year after a change in short-term interest rates. After the first few years, the interest sensitivity of aggregate output to a change in short-term rates is smaller today than it was in the earlier decades. The changes in the response of aggregate output to changes in interest rates mask some larger, but mostly offsetting, changes in the responses of different sectors. Both residential and nonresidential construction are less sensitive to interest rates. In residential construction, the reduction in sensitivity reflects the absence of disintermediation-induced episodes of credit rationing (disruptions in the supply of credit that occurred when funds dried up during periods of high interest rates); it does not reflect any reduction in the direct effect of interest rates on the demand for housing. The traded goods sector has become more responsive to changes in interest rates because the exchange rate has become more sensitive to the difference between U.S. and foreign interest rates. Monetary policy affects consumption spending and investment in producers' durable equipment much the same as before. In financial markets, long-term interest rates appear to have responded more quickly in the 1980s than they did before to changes in short-term rates.


In the MPS model, the structure of the monetary transmission mechanism draws on two critical characteristics of the general "Keynesian" paradigm. First, changes in the supply of real money balances affect spending only through changes in interest rates. There is no direct channel from a change in money balances to spending. Second, these changes in interest rates generally imply changes in real rates in the short run because wage and price expectations adjust only sluggishly. This article thus identifies the stance of monetary policy with movements in short-term interest rates-specifically the overnight federal funds rate, which is commonly regarded to be controllable by Federal Reserve actions.

In the MPS model, changes in the central bank's monetary stance affect spending and output directly through three primary channels: the influence of the cost of borrowed funds on business and household investment decisions; the influence of the value of wealth on consumption; and the influence of the exchange rate on the volume of imports and exports. Although these channels are not independent of one another, the direct quantitative importance of each can be gauged by tracing the effects of a 1 percentage point reduction in the federal funds rate on spending when each channel operates alone. To emphasize the interest sensitivity of each component of spending, no feedback is allowed from one sector's spending to another sector's spending. In addition, wages and prices are held fixed so that the change in the federal funds rate is a change of the same magnitude in the real rate of interest.

The influence of the federal funds rate on the cost of capital and through the cost of capital on consumption and investment is the largest of the three channels and accounts, on average, for about 55 percent of the total direct effect. The cost of capital is a measure of the rate of return on an investment that is necessary to cover the costs of financing, depreciation, and taxes. Financing costs, in turn, consist of the interest paid on bank loans or marketable debt and the costs of raising funds in the equity market. In some instances, such as inventory investment, spending behavior appears to be sensitive to movements in short-term interest rates. For longer-lived investment, such as business fixed investment or residential construction, long-term interest rates are more relevant to the investment decision.

Variations in the cost of capital alter the desired proportions of capital and labor in production, the desired stock of housing relative to income, and the desired stock of consumer durables relative to income. A decline in the cost of capital increases the desired stock of capital, and spending consequently increases relative to a baseline of higher interest rates--initially to obtain the additional capital and then to offset the depreciation on the larger stock of capital. The cost of capital has its largest effects on business fixed investment and residential construction activity. The pattern of cost-of-capital effects over time reflects the lag between a change in the federal funds rate and changes in long-term interest rates and also between changes in these interest rates and changes in investment.

In terms of the direct effect over five years, a change in the stock of wealth is next in importance in transmitting changes in the federal funds rate to changes in spending: It contributes on average 28 percent of the total direct effects. In the MPS model, a change in wealth directly affects only consumption spending. The timing of the effect of changes in wealth on consumption shown in table 2 takes into account the lag between changes in the federal funds rate and changes in the corporate bond rate, the yield on equity, and the price of land. (3)

Finally, a decline in U.S. interest rates leads to a depreciation of the dollar, which boosts net exports as domestically produced goods become more competitive. This channel contributes, on average, 17 percent of the total direct effects. The magnitude and the timing of the response of net exports reflect the estimated lag between changes in the exchange rate and changes in import and export prices and between changes in these prices and changes in the demand for imports and exports. The calculated contribution of this channel is based on the assumption that foreign interest rates do not respond to the decline in U.S. interest rates, and thus it tends to overstate the exchange rate influence should foreign monetary authorities correspondingly reduce foreign interest rates.

Comparing the estimates in tables 1 and 2 [omitted] with calculations DeLeeuw and Gramlich made more than twenty years ago using an early version of the MPS model provides an interesting historical perspective. (4) As they are in the current structure of the model, the cost of capital and the value of wealth were of greatest importance for the transmission of monetary policy. But the link between credit rationing and residential construction, rather than variations in the exchange rate, was identified as the third channel of transmission. This difference reflects some of the structural changes described at the beginning of this article. In the 1960s, variations in the exchange rate were infrequent because exchange rates were, for the most part, fixed under the Bretton Woods agreement. And mortgage credit and, consequently, housing activity were at times constrained by the regulation of interest rates on savings deposits and the absence of readily available alternative sources of mortgage funds.

Although the results presented in tables 1 and 2 [omitted] suggest that monetary policy exerts a powerful influence on real output, the simulation exaggerates the actual influence on the sustainable level of real output because it abstracts from two critical aspects of the economy. First, the simulation holds wages and prices constant and thus creates the impression that the supply of labor and the production of output adjust fully to the increase in demand generated by the decline in interest rates. In fact, wages and prices appear to be sufficiently flexible--albeit with some lag--so that, as aggregate demand increases and labor markets tighten, the burden of the adjustment will shift to wages and prices. In the long run, changes in the level of the money stock cause changes only in the price level and are neutral with respect to the level of real production, which is constrained by the quantities of labor and capital available, the production technology, and the rate of technological progress.

A second, related simplification is the assumption that monetary policy can peg the rate of interest for an extended period. An attempt to conduct monetary policy by permanently lowering the nominal rate of interest could eventually lead to an unstable path for real output and prices. Lowering the nominal rate of interest lowers real interest rates initially by a similar amount and leads to an increase in output, a reduction in unemployment, and an increase in inflation. But this last effect further reduces real interest rates and consequently raises output and inflation further. By holding prices fixed, the simulation reported in tables 1 and 2 hides the potential instability created by a policy that pegs nominal interest rates.

The lower panel of table 3 presents a simulation of the full model, which drops both of the simplifying assumptions made earlier. (5) A comparison of the upper and lower panels illustrates the transitory effects on real output of a change in the stance of monetary policy when wages and prices are allowed to respond fully. The easing of the monetary stance stimulates demand pressures that cause a small rise in prices within one year and larger increases subsequently. The simulated path of real output is cyclical, but the effect on activity is negligible in the long run.


In general, the equations of the MPS model have been estimated under the assumption that, aside from the absence after 1980 of credit-rationing constraints on housing expenditures, no significant shifts have occurred in economic behavior over the estimation period of nearly thirty years. By examining in more detail the links among money, interest rates, and asset values and applying statistical tests aimed at identifying shifts in these relationships during the past ten years, this section weighs the validity of that assumption.

From Money to Short-Term Interest Rates: The Demand for Money

A stable empirical relationship between money and market interest rates has been difficult to pin down since the early 1970s. In hindsight, most of this instability can be accounted for either by innovations in financial practices designed to raise the effective rate of return on money balances or by the subsequent removal of regulations that constrained the rate of return on deposits. The creation of new depository instruments further weakened the reliability of the relation between money and interest rates.

The link between money and short-term interest rates is usually based on some theory of the transactions demand for money. For a long time, a simple model of the demand for money-popularized by William J. Baumol and James Tobin and based on models of optimal inventory behavior-was used in the MPS model and many of the other large-scale macroeconomic models to describe the behavior of M1 demand. The model tended to fit the data quite well. According to this theory, the optimal level of money balances was determined by offsetting the interest forgone in the holding of non-interest-bearing money, instead of interest-bearing but less liquid assets, with the lower transaction costs resulting from less frequent switching between money and interest-bearing assets. The specification led to an inverse relation between the short-term market rate of interest and the quantity of money demanded, and to a positive relation between the level of nominal income and the quantity of money demanded.

By 1974, the fit of these equations began to deteriorate. The actual level of M1 (specifically, the demand deposit component of M1) fell far short of the quantity predicted by the equation. It is now clear that the shortfall in money growth was related to various innovations in financial markets, some designed to reduce the transactions costs of switching between money and an interest-bearing asset and others to reduce the variance and uncertainty of cash flow. The introduction and growing use of many of these techniques (for example, overdraft accounts, repurchase agreements, and remote disbursement of checks) was apparently related to the spread of computer technology in the banking industry starting in the mid-1970s. These innovations ultimately reduced the demand for the traditional medium of exchange. (For more information on this subject and others discussed in this article, see the selected reading list at the end of the article.)

At the same time, changes in regulations expanded the range of instruments that could serve as a medium of exchange, one of the more far-reaching of which was an interest-bearing checkable account for households, state and local governments, and nonprofit institutions. The increasing use of these accounts led the Federal Reserve in 1980 to distinguish MI, naming it MIA, from a slightly broader aggregate, MIB, which included these other checkable deposits (OCDs). Although the aim in creating MIB (now referred to as MI) was to construct an aggregate with a predictable and stable relationship to income and interest rates, the empirical stability of MIB was short-lived. The demand deposit component of MIB became increasingly difficult to explain. Moreover, the emergence of two instruments that were close substitutes for the OCD component of MIB--money market mutual funds in the late 1970s and money market deposit accounts in 1982-argued for emphasizing a larger aggregate that encompassed them. Thus, over the past several years, attention has shifted to M2, where some success has been achieved in modeling aggregate M2 as a function of nominal income or consumption and the opportunity cost, defined as the excess of market rates of interest over the rates paid on M2 components. Because most M2 components bear interest rates that respond gradually to the return on market instruments such as Treasury bills, M2 is more responsive in the short run than in the long run to movements in market interest rates. Modeling the sluggish behavior of M2 rates appears to be important in specifying a stable equation for M2 demand.

The behavior of the monetary aggregates since the mid-1970s points to changes in and greater uncertainty about the relations between money and income and between money and the rate of interest. it thus has reduced the indicator value of money in gauging the present state of the economy and diminished the usefulness of money as an intermediate target in the presence of uncertainties about the economy. (6) From the perspective of the model, however, uncertainty about the demand for money is of limited practical import. The monetary authority remains capable of influencing output in the short run as long as it can directly influence the short-term interest rate. The instability of the money demand function just makes more difficult the task of estimating the change in the monetary aggregate that would accompany any given change in the short-term interest rate.

From Short-Term to Long-Term interest Rates: The Term Structure

In the model, short-term interest rates--for example, the commercial paper rate--are important in explaining inventory behavior and expenditures on consumer durables. They are less important in explaining residential construction and business fixed investment, which are assumed to be more sensitive to longer-term rates of interest. Long-term interest rates are also important in the transmission process because of their role in explaining variations in wealth, itself a key determinant of consumption.

A standard view of the determination of long-term interest rates is that arbitrage between short-term and long-term debt instruments should equalize expected returns over common holding periods, except possibly for risk premiums. This view implies that the yield on a long-term asset, such as a corporate bond, should be a weighted average of the current and expected future short-term rates over the life of the asset. In the MPS equation, the determinants of the expected path of future short-term interest rates are assumed to be past values of the commercial paper rate and of the inflation rate. This specification, based on the work of Franco Modigliani and Robert J. Shiller, assumes that changes in the commercial paper rate ultimately are reflected in entirety in changes in the bond rate and changes in the inflation rate have only a transitory effect on the bond rate. (7) The pattern of lag coefficients is permitted to vary with the level of recent short-term rates to reflect the dependence of the duration of coupon bonds, such as corporate bonds, on the interest rate. (8)

The institutional changes noted earlier-in foreign exchange markets, in housing finance, and in household and corporate balance sheets--did not in themselves necessitate any change in the behavior of the corporate bond rate in the 1980s. Nevertheless, the relation between the bond rate and its determinants may have changed because the assumptions underlying the formation of expectations of future short-term interest rates--that they are based on past short-term interest rates and inflation rates--might not be appropriate in all circumstances. A popular alternative view of the expectations process--particularly compelling in the description of financial markets--emphasizes the forward-looking nature of expectations, in which informed judgments about the future course of events are based on knowledge of the structure of the economy, including the process whereby economic policy is formed. Although such "rational" expectations may at times be consistent with using past values of a particular variable to forecast its future values, the conditions under which this procedure is optimal are relatively restrictive.

The MPS equation for the corporate bond rate is subjected to a standard statistical test to determine whether the relation between the corporate bond rate and past short-term interest rates has remained constant over time. This test allows the pattern of coefficients on the paper rate to vary between the 1960s and 1970s as compared with the 1980s but preserves the long-run properties of the equation. Thus, changes in the commercial paper rate are ultimately reflected in entirety in the bond rate.

The result of this test suggests that the behavior of the bond rate in relation to its determinants has changed over the 1980s and that the change has been in the direction of shortening the lag between changes in the short-term interest rate and changes in the bond rate (table 4, line 1). In the 1960s and 1970s, a rise of I percentage point in the commercial paper rate raises the bond rate by 0.21 percentage point in the same quarter and by 0.36 percentage point within one year; in the 1980s, the corresponding responses are 0.32 percentage point in the same quarter and 0.46 within the year. (9) These results are consistent with the hypothesis that investors believe the current short-term rate to be a better predictor of future short-term rates than it was earlier and therefore link the long-term rate to the current short-term rate more quickly than previously. Although the faster passthrough of a change in the short-term rate to a change in the long-term rate is significant statistically, the change is too small to have much effect on output in the short run.

From Long-Term Interest Rates to the Return on Equity.

The relation between the return on equity and the return on debt also is based on the assumption that arbitrage equalizes the expected real rates of return on the two, apart from a possible risk premium. Arbitrage implies that the price of equity is based on investors' expectations about the stream of future dividends payable to shareholders and on the rate of return to investments that are alternatives to equity.

In the MPS equation, the real rate of return on equity is represented by the dividend-price ratio, and the expected real rate of return on long-term debt is represented by recent rates of return on corporate bonds less the inflation rate averaged over a long period. (10) A priori, the estimated coefficients on the bond rate should sum to 1.0, and the estimated coefficients on the inflation rate should sum to - 1.0, so that a rise in the nominal interest rate due to an increase in the rate of inflation is correctly perceived to leave the real rate of return on bonds unchanged. The estimated coefficients in the current version of the equation satisfy these a priori values. (11) However, the current version was estimated using data only through 1973, and extensions of the estimation period through the 1970s or 1980s result in very different estimates for the coefficients on the inflation rate. As Franco Modigliani and Richard A. Cohn first noted, the coefficient sum on the inflation rate falls dramatically over longer sample periods and is not statistically different from zero. (12) Thus, equations estimated through the 1970s or 1980s that show the sum of coefficients on the bond rate remaining close to 1.0 suggest that investors equate the real return on equity with the nominal return on debt. (13)

The test for structural shift is reported with the caveat that there is some risk of mistaking structural change for misspecification in this case because the estimated equation over the extended period does not appear to be consistent with rational behavior. (14) The test results suggest rejecting the hypothesis that, at standard statistical significance levels, the response of the return on equity to changes in the bond rate is the same in the 1980s as it was in the 1960s and 1970s. Although the estimated coefficient sums on the bond rate are similar across periods, the individual coefficients indicate some slowing in the response of the dividend-price ratio to the bond rate over the 1980s. For the 1980s, the contemporaneous coefficient on the bond rate is 0.38, compared with 0.52 for the estimation period ending in 1979. (15) Other things being equal, this result would reduce the effect of monetary policy in the very short run. However, as was true for the behavior of the corporate bond rate, the size of the change is such that its effect on spending would be negligible in the short run. And the overall response suggests that changes in bond rates continue to influence the return in the equity market in a critical way.

From interest Rates to Exchange Rates Models of the determination of the exchange rate unfortunately have not been particularly successful in explaining movements of exchange rates since the floating of the dollar in 1973. Equations that fit well over some estimation period often perform poorly over the postestimation period. Given the rapid change in the integration of international capital markets, this lack of success may not be surprising.

In the MPS model, the determination of exchange rates is based on the portfolio balance theory. According to this theory, demands of investors for assets denominated in different currencies depend, aside from possible risk premiums, on differences in expected returns. Comparing returns on such assets requires that one of the assets be converted to the currency of the other. Thus, to compare dollar- and yen-denominated assets, for example, an investor must know the rates of return offered on each and must have some expectation of the change in the dollar-yen exchange rate over the holding period. In the special case in which investors consider assets denominated in different currencies to be perfect substitutes, that is, investors are indifferent to risk, expected holding-period returns-converted to a common currency--are equalized.

The extent to which domestic monetary policy affects exchange rates--and through exchange rates, the volume of imports and exports--depends on how sensitive investors' desired portfolio allocations are to the difference between expected holding-period yields. The greater is the sensitivity of demands--that is, the more perfect the substitution between foreign and domestic assets--the greater will be the change in the spot value of the domestic currency for a given change in the domestic interest rate, other things being equal.

To see this, consider what happens to the demand for domestic currency in the foreign exchange market after an increase in the domestic interest rate. For given values of foreign interest rates, exchange rates, and expected future exchange rates, domestic assets are now more attractive than they were before the rise in the domestic interest rate, and the demand for assets denominated in the domestic currency increases as holders of foreign assets try to switch into domestic assets. But if trade in goods and services is unresponsive in the short run to movements in spot exchange rates, the supply of domestic-currency assets in the foreign exchange market does not increase, and the prices of domestic currency in terms of foreign currencies (exchange rates) must rise to reestablish equilibrium in the foreign exchange market. The more responsive asset demands are to interest rate differentials, the greater will be the initial increase in the demand for domestic currency and the larger the consequent appreciation of domestic currency that is needed to equilibrate demand and supply of foreign exchange. From the perspective of asset holders, the appreciation of the domestic currency sets up the expectation of a future depreciation of domestic currency (because the expected level of exchange rates is so far unchanged) and has the effect of raising the expected return on foreign assets or, equivalently, of reducing the expected return on domestic assets.

The two equations in the MPS model explaining the U.S. demand for foreign assets and the foreign demand for U.S. assets were estimated using data through the mid-1970s and are fairly conventional: Demands for assets are related to the size of the economy, to the domestic interest rate, to a trade-weighted average of foreign interest rates, and to various proxies for the expected change in a trade-weighted average of the foreign exchange value of the dollar. A sense of the nature of any structural change can be gained from examining the exchange rate errors produced by these equations. (16) These errors indicate a sizable underprediction of both the level and the variability of the exchange rate over the 1980s. The positive correlation of the errors (defined as the actual exchange rate less the predicted exchange rate) with the size of the interest rate differential (defined as the domestic interest rate less the foreign interest rate) suggests that the asset equations may understate the interest sensitivity of asset demands, especially over the 1980s. One way of attempting to quantify the structural change is to raise the coefficients in the asset demand equations that represent the sensitivity of asset demands to the interest differential by enough to eliminate the correlation between the exchange rate errors and the interest differential. (17) The resulting coefficients on the interest rate differential are 60 percent larger over the 1980s than over the 1970s. Thus, the sensitivity of exchange rates to interest rates appears to have increased significantly in the 1980s, a finding that is consistent with views that world capital markets have become more closely integrated. For any given change to the differential between U.S. and foreign interest rates, the change to spot exchange rates necessary to reequilibrate the balance of payments is larger compared with that in the 1970s.


The efficacy of the monetary transmission mechanism requires not only that short-term interest rates influence long-term rates and asset prices but also that subsequent changes in the returns on financial assets, in turn, influence spending. Whereas the preceding section examined the behavior of financial markets and asset prices and identified, among other things, an increased interest sensitivity of the exchange rate over recent years, this section examines the relation between rates of return on financial assets and spending. In so doing, it shows that the sensitivity of spending to financial variables appears to have decreased somewhat over recent years.

Residential Construction

The component of aggregate demand most commonly perceived to have undergone major regulatory and structural changes over the past decade is expenditure for residential construction. These changes are frequently cited as having reduced the interest sensitivity of housing expenditure and, therefore, the likelihood that this sector will bear a disproportionate share of the effect on real spending of monetary tightening. First, several innovations and regulatory changes have mitigated the tendency for financial intermediaries to experience deposit outflows during a period of high interest rates. In 1978, six-month money market certificates with rates tied to Treasury bill yields were authorized; and in 1979, small-saver certificates with no minimum denomination and with yields based on Treasury securities were introduced. Beginning in 1980 and continuing for several years, a sequence of regulatory acts lowered withdrawal penalties on time deposits and phased out ceilings on deposit rates.

Along with these changes came the growth in the market for mortgage-backed securities, whereby mortgage-originating institutions pooled and resold their mortgages to third parties. As a share of residential mortgage debt outstanding, mortgage pools backing securities grew from 1 percent in 1970 to 12 percent in 1980 and to 35 percent in 1989. (18) By broadening the base of potential credit suppliers, the securitization of mortgages weakened the link between deposit growth at savings institutions and the availability of new mortgages. This development, with the regulatory changes affecting deposit rates, reduced nonprice constraints on the supply of mortgage funds, which had been associated with disintermediation at thrift institutions.

A third development in the mortgage market--the emergence and increased use of adjustable-rate mortgages (ARMs)--may have further altered the sensitivity of expenditures on residential construction to changes in interest rates. Two factors are often cited in support of this argument. First, home buyers who might have postponed purchases when they thought fixed-rate mortgage terms were unusually high relative to short-term interest rates would now be less likely to do so because ARMs are typically indexed to short-term interest rates. In fact, this factor may have little effect on the long-run interest sensitivity of housing demand, although it may reduce the effect in the short run. Second, compared with a fixed-rate mortgage, an ARM may offer a significant reduction in financing costs for those buyers whose intended holding period is relatively short, because of the common practices of offering an initial-year discount and capping the size of yearly changes permitted in the ARM rate.

Tests of the hypothesis that the interest sensitivity of housing demand has changed are carried out with the MPS equation for residential construction. This equation relates the desired stock of housing to a proxy for permanent income and to the cost of capital for housing and explains actual construction expenditures as adjusting with some lag to the desired housing stock. (19) Estimates indicate that expenditures also depend on the change in the unemployment rate and have been constrained at various times by credit rationing in the mortgage market. (20)

The residential construction equation is reestimated so as to permit the intercept and the sensitivity of the desired housing stock to the cost of capital to differ between the 1960s and 1970s on the one hand and the 1980s on the other hand. (21) Although the estimated coefficient on the cost of capital is smaller over the 1980s than over the previous two decades, the difference is not statistically significant. (22)

Although the demand for housing appears to have been no less sensitive to the cost of capital over the 1980s, expenditures on residential construction may nevertheless have been less interest sensitive because disintermediation-induced episodes of credit rationing were absent during the 1980s. To quantify the total effect of interest rates on housing expenditures-including that occurring through credit rationing in the 1960s and 1970s--the equation is reestimated without the proxies for credit rationing, while the estimated intercepts and coefficients on the cost of capital are still allowed to differ for the two subperiods. Because of the positive association between credit rationing and interest rates, the estimated sensitivity to the cost of capital over the earlier subperiod increases without these proxies. When credit availability proxies are excluded, the effect on housing expenditures in the first year after a change of 1 percentage point in the cost of capital (due, say, to a change of 1 percentage point in the after-tax mortgage rate) increases from 16 percent to 21 percent. (23) By contrast, the estimated response of residential construction over the 1980s to the same change in the cost of capital is about 9 percent. (24)

Nonresidentail Construction Expenditures

Despite significant changes during the first half of the 1980s both in the real rate of interest and in aspects of the tax code pertaining to structures, the MPS equation successfully explains construction spending through 1985. (25) The form of the equation follows the neoclassical theory of investment, which is that capital is substitutable to some degree for labor, so that the desired capital intensity of production is inversely related to the cost of capital relative to the cost of labor. The higher are the real financing costs--either the rate of interest on debt or the rate of return on equity--the lower will be the desired stock of capital relative to output. In the current version of the model equation, the long-run degree (or elasticity) of substitution is 0.5, so that a 1.0 percent change in the wage rate relative to the cost of capital will change the long-run ratio of capital to output by 0.5 percent. (26)

A breakdown occurs after 1985 in the relation between investment and its determinants; it is highlighted by the plots in chart 1 [omitted] of actual and predicted values of construction expenditures where the predicted values are based on an equation estimated through 1979. The estimated equation captures the movements in construction spending quite well through 1985. After 1985, when construction spending stagnates, the predictive performance of the equation deteriorates, and overprediction errors occur--of almost 18 percent in 1986 and of about 48 percent in 1988.

Tests for changes in the short- and intermediate-run interest sensitivity of construction are not definitive when applied to this equation because the interest rate does not enter independently of output: The explanatory variables are specified as the product of output and the capital-output ratio (which, in turn, depends on the rate of interest). Thus, a test for possible structural shift since 1986 in the sensitivity of construction to interest rates cannot be performed independently of a test for structural shift in the sensitivity to output. Applying the conventional test without distinguishing between shifts in these sensitivities generates a probability less than 1 percent that the response of nonresidential construction expenditures to changes in output or in the cost of capital is the same after 1985 as before. (27)

The behavior of construction expenditures after 1985 suggests a reduced sensitivity to the cost of capital and a faster adjustment to changes in output. Two pieces of evidence support this observation. First, for equations estimated through 1985, the post-1985 prediction errors are smallest when the relative cost of capital is held constant at its 1985 value. Holding the relative cost of capital constant is essentially equivalent to assuming that capital and labor are not substitutes for one another in production. Second, to explain the behavior of construction only over the period from 1986 to 1989, one does best by modeling investment as adjusting in proportion to the change in output and as not responding at all to changes in the relative cost of capital. Moreover, if the 1986-89 estimated speed at which investment adjusts to output is compared with the pre-1986 estimate, one finds that the adjustment speed has increased although it may be imprecisely estimated given the short estimation period.

Several reasons have been offered for the apparent lessening of the importance of the cost of capital for post-1985 investment behavior. These reasons are based on a more eclectic view of the determinants of investment than that inherent in the standard neoclassical approach. One argument attributes the investment collapse to high vacancy rates. High vacancy rates could matter if, by signaling losses in the construction industry, they provide information on expected output or the cost of capital not captured by simple distributed lags on these terms. However, high vacancy rates do not provide a clear-cut signal: Although they may suggest an untenable position in the long run, they may not mean that the owners of the buildings are losing money in the short run. (28)

Another explanation of investment behavior after 1985 is associated with lending institutions' eagerness through the middle of the decade to support construction activity and their declining interest in providing these funds more recently. The pattern of errors from the model equation does not support this explanation: Actual construction expenditures should have exceeded predicted expenditures when funds were readily available, but they did not. However, the absence of underprediction errors in the earlier part of the 1980s could reflect offsetting errors in the model specification. Similarly, because actual construction during this period was not significantly stronger than predicted construction, the equation errors also fail to support theories that foreign money was more readily available or that business sentiment was more bullish earlier in the decade.

The standard view of investment spending--that the parameters of the production process along with the relative costs of capital and labor determine the optimal capital stock-may be basically correct. Even so, the poor performance of the equation after 1985 may reflect misspecification in the cost of capital either because the model does not use an appropriate proxy for expectations of the real return on debt and equity or because it does not accurately calculate the effects of the change in the tax law. In any case, the bottom line suggests either serious misspecification in the cost of capital or the presence of some factor that reduced the apparent sensitivity of construction spending to the cost of capital. If the latter is true, this factor has not yet been identified.

Investment in Producers' Durable Equipment

Outlays for capital equipment are typically considered one of the areas of spending most sensitive to changes in interest rates. This belief appears to rest on the notion that a high degree of substitution between capital and labor is possible in production. In the MPS equation for investment in equipment and machinery, the estimated long-run elasticity of substitution is not significantly different from 1.0, so that a 1 percent increase in the cost of capital relative to the cost of labor ultimately leads to a 1 percent decline in the ratio of equipment to output. In this regard, the stock of equipment may be thought to be more interest sensitive than is the stock of structures. Nevertheless, changes in interest rates have a smaller effect on the cost of short-lived capital, such as equipment, than they do on the cost of long-lived capital. For instance, when the life span of capital is five years (a 20 percent depreciation rate), an increase in the real financing costs from 3 to 4 percentage points raises the cost of capital from 23 percent to 24 percent. By contrast, when the lifespan is twenty years (a 5 percent depreciation rate), the same rise in the real financing costs raises the cost of capital from 8 percent to 9 percent--a much larger increase in percentage terms.

The importance of the life span of capital in determining capital's sensitivity to interest rates has been cited as a potential source of change in the interest sensitivity of investment because the growing share of equipment accounted for by computers and high-technology equipment has reduced the expected life span of this aggregate. In 1970, the depreciation rate for the stock of equipment was 12.5 percent a year (assuming an exponential pattern of depreciation); in 1990, it was over 15 percent a year. Moreover, investment in computers may be more sensitive to product cycles than to financing costs or to expected output. In contrast to these factors, an increase in the interest sensitivity of aggregate investment might be expected if higher leverage ratios increased the interest sensitivity of a firm's cash flow.

Neither of these developments lends itself to a definitive test for changes in interest sensitivity, given the specification of the MPS equation for investment. In the first instance, the model ties down a priori the long-run response of investment to a change in the real rate of interest: The capital stock and, consequently, long-run investment rise by 1 percent for a 1 percent decline in the cost of capital (based on an estimated elasticity of substitution equal to one); and the percentage decline in the cost of capital for a given percentage decline in the rate of interest varies inversely with the depreciation rate of capital. Thus, as the life span of the aggregate capital stock declines (the depreciation rate rises), the elasticity of long-run investment with respect to the rate of interest declines. (29) Although the appropriateness of a long-run unitary elasticity of substitution for computers in the production process theoretically can be tested, it is not tested because the technique used to estimate this elasticity is imprecise and because the estimate may be influenced by the high growth rate of computer investment. The test for structural change thus allows only for a change in the short- and intermediate-run sensitivity of investment to the rate of interest.

However, as was true for nonresidential structures, a test for structural change, based on the MPS equation, cannot distinguish between change in the short- and intermediate-run sensitivity to interest rates and change in the sensitivity to output. The test requires that the determinants of spending enter independently of one another, a condition not satisfied by the equation specification. Testing for a shift in the overall response of investment to its determinants--the cost of capital and output--suggests only a 3 percent probability that there has been no significant change.

Whereas the statistical evidence favors change, other evidence suggests that the implied change is marginal, especially as it affects the interest sensitivity of investment. First, the post-1980 predictions from an equation estimated from 1961 through 1979 are almost identical to the predictions from an equation estimated from 1961 through 1987. Second, when the equations for the two periods are simulated after a change in interest rates, the patterns of investment are similar. For a 1 percentage point change in the interest rate, the maximum difference at any time between the two predicted paths is $1 billion. That is, for approximately a 5 percent change in the desired capital stock at current rates of interest and depreciation, the maximum difference is about 8 percent of the long-run response. Third, more sizable differences occur between the patterns of investment short-lived capital than in the stock of long-lived capital, replacement investment associated with a unit of short-lived capital is greater than that associated with a unit of long-lived capital. For normal ranges of real interest rates, growth rates, and equipment depreciation rates, the response of investment to a change in the interest rate declines as the life span of equipment shortens. in response to a change in output. These three results suggest that a shift in the response of investment to output, rather than in the response to interest rates, explains the low statistical probability of an unchanged response of investment to its determinants.

Consumption of Nondurable Goods and Services

Household expenditures for services and nondurable goods form the largest component of final demand. Thus, any change in the sensitivity of consumption to monetary policy will have important repercussions for the sensitivity of aggregate output to interest rates.

One argument for a change in the interest sensitivity of consumption is rooted in the idea that, since the deregulation of deposit rates, household interest income has been linked more closely to movements in interest rates. If households make spending decisions based on current income because, say, they are unable to borrow against future income (that is, they are liquidity constrained), greater interest sensitivity of income will induce greater interest sensitivity of consumption (where a rise in rates causes a rise in consumption). If spending depends on average income rather than on current income, however, the increase in the interest sensitivity of consumption will be more moderate.

Another argument for a changed interest sensitivity of consumption is that the growth in the share of adjustable-rate mortgages and home equity loans in mortgage debt raises the probability that households will incur liquidity problems when interest rates rise. In this case, however, a rise in interest rates will reduce consumption. An implicit assumption in this scenario is that households with adjustable-rate debt do not have rate-sensitive assets with rising returns to offset the increase in debt-servicing costs.

The MPS equation for consumption of nondurable goods and services was used to identify changes in consumption behavior that might reflect the aforementioned changes in the interest sensitivity of household cash flow. The equation is based on the life-cycle model of consumption--which assumes that households may borrow against future income and thus consume in excess of current income and wealth; still, the specification could support an important role for liquidity constraints depending on the estimated coefficients on current and recent income relative to those on more distant income and wealth. Estimates for both subperiods, while statistically significant for current and recent income, suggest that distant income and wealth are also important in determining consumption behavior. Thus, even if household interest income or household interest payments are more interest sensitive now than they were earlier, the quantitative effect on the interest sensitivity of consumption may be small. (30)

The influence of monetary policy on consumption can change also if the relation of consumption to wealth and property income has changed. These are the determinants of consumption most closely linked to changes in interest rates. (31) In the estimation of the consumption equation, when the response to these determinants is allowed to change, the data strongly support the existence of a stable relation over time.

Given these findings--that the wealth channel has been stable over time and that households appear not to be unduly liquidity constrained--any change in the relation between short-term interest rates and consumption rests upon a change in the effect that short-term rates have on the market values of corporate equity and of noncorporate land. (In the MPS model, the market values of other components of household wealth--bonds, consumer durables, and housing--are not tied directly to changes in interest rates.) But changes in the relation between short-term interest rates and corporate equity are small because of offsetting changes in the speeds at which short-term rates pass through to long-term rates and at which long-term rates pass through to equity prices. Also, changes in the relation between short-term interest rates and noncorporate land values are quantitatively insignificant because of the slowness with which land values are estimated to adjust to changes in interest rates.

Consumer Durable Expenditure

In the MPS model, the determinants of the desired stock of consumer durables are income, consumption of nondurable goods and services (a proxy for permanent income), and the cost of durable goods relative to that of nondurable goods. The cost of a durable good takes into account the financing and depreciation expenses associated with the use of the good and is, in this sense, akin to the cost of business capital. Variations in interest rates affect the relative cost of consumer durables and therefore the demand for them. By measuring the responsiveness of purchases of durable goods to changes in the relative cost of durables, one can determine whether or not this particular channel for monetary policy has changed. (32) The equations relating consumer purchases of motor vehicles and other durables were reestimated allowing the response to the relative cost of durable goods to differ over the pre- and post-1980 periods. The results strongly favor the hypothesis that the response of purchases to the cost of capital for consumer durables has been stable over time.

Business Inventory Investment

As with consumer purchases of durables, monetary policy and inventory investment are directly linked through the cost of capital. The higher are the costs of financing inventories, the lower will be the desired stock of inventories relative to sales. In estimation, the sensitivity of the inventory-sales ratio to the cost of capital is generally small but statistically significant. When the behavior of inventories is examined over the pre-and post- 1980 periods, the response to variations in the cost of capital generally appears to be stable. The estimated sensitivity of nonretail durable inventories to the cost of capital has increased a bit, although the increase is not statistically significant at the normal levels of significance.


The preceding two sections have detailed the particular changes in the interest rate sensitivity of asset prices and spending components that have occurred since 1980. In this section, the MPS model is used to quantify the net effect on the economy of the changes found in the response of asset prices to short-term interest rates and in the response of spending to interest rates and asset prices. Changes in the behavior of financial markets and asset prices are most notable in the increased sensitivity of exchange rates to the differential between yields on domestic assets and those on foreign assets. The short-run efficacy of monetary policy is diminished slightly by the slower passthrough of changes in bond rates to the yield on equity, but this diminution is largely offset by a faster passthrough of changes in short-term rates to long-term rates.

Significant changes to the interest sensitivity of spending have occurred in residential and nonresidential construction expenditures. In both cases, the interest sensitivity has fallen. Institutional and regulatory changes beginning in the late 1970s have eliminated the episodes of credit rationing that constrained housing construction during times of high interest rates in the 1960s and 1970s. No satisfactory explanation has been offered for the apparent insensitivity of nonresidential construction to the cost of capital since around 1986.

To quantify the net effect of these changes on the ways in which monetary policy influences the economy and to assess the overall effect, the partial and full model simulations presented in tables 1 and 2 [omitted] are repeated. With the same baseline, the simulations first use the equations that best represent the pre-1980s behavior of asset prices and spending and then use the equations that best represent post-1980s behavior. The equations for asset prices and for spending differ in the two versions of the model only when their estimated sensitivity to interest rates has changed significantly over time. (33)

Table 5 [omitted] presents the effect on real spending in the two versions of the model of a I percentage point reduction in the federal funds rate, when multiplier and accelerator feedbacks to spending, and output feedbacks to interest rates and prices, are suppressed. As was done in tables 1 and 2, the separate contributions of the three principal channels for monetary policy are quantified.

The results suggest that the direct effects of changes in the federal funds rate on real spending are about the same before and after 1980 for the first two to three years but are somewhat smaller after 1980 during the fourth and fifth years. The principal reason for the apparent reduction over the 1980s in the direct effect of interest rates on spending in the intermediate run is the absence of credit rationing. Previously, the effect that credit rationing had on residential expenditure was equated with a doubling of the sensitivity of the desired housing stock to the cost of capital. That is, in the first year after an increase of 1 percentage point in the mortgage rate, residential construction would decline about 9 percent after 1980 and about 21 percent before 1980. At 1990 levels of residential construction, this difference is about $20 billion dollars, a sizable portion of the difference in aggregate output generated by the two versions of the model. Of secondary importance to the diminished response to interest rates is the failure of nonresidential construction spending to respond to variations in the cost of capital after 1985. Table 5 [omitted] also highlights the increase in the importance of the exchange rate channel for the transmission of monetary policy in the post-1980 world as represented by the MPS model. The changes to the asset demand equations are such that the relative contribution of the exchange rate channel has doubled on average since 1980. The relative importance of the wealth channel is about the same in the pre- and posts 1980 versions of the model.

Both versions of the model are simulated again, allowing all sectors of the model to respond and assuming a phased increase of 1 1/2 percent in the level of M2. As in the partial model simulations, the effects on real GNP are similar in the two versions for the first few years. But, in contrast with the direct effects on spending as demonstrated by the partial simulations of table 5 [omitted], the difference between the two versions in the effect on aggregate output diminishes over time because in both versions of the model long-run output depends only on the production technology and the supply of capital and labor. In the long run, the level of money has no effect on output.


Structural change, whether due to changing markets, altered institutions, or technological innovations, is difficult to capture in macroeconomic models. The approach discussed in the body of this article is to test for one-time shifts in coefficients of the MPS model. The assumptions of the one-time-shift analysis are that selected coefficients of the model may have shifted around 1980 and that these shifts will persist indefinitely. (34)

An alternative method of analyzing structural change, outlined in this appendix, permits model coefficients to vary continuously over time. The time-varying-coefficient analysis employs the less-restrictive assumptions that the timing of structural change is generally unpredictable and that structural change may be intrinsically evolutionary, with some coefficients subject to continuous changes and others to infrequent changes. This analysis assumes also that coefficient changes may be partly transient and may provide only partial predictions of future values of the model coefficients.

Under these more general assumptions, the time-varying-coefficient analysis can provide additional evidence on the following three issues regarding structural change. First, what proportion of the total variation in variables explained by model equations can be ascribed to estimated changes in model coefficients, especially those coefficients associated with key channels of policy transmission? Second, are the estimated changes in model coefficients larger in the 1980s than in earlier periods? Third, to what extent are changes in model coefficients persistent or transitory?

In brief, the results of the time-varying-coefficient analysis suggest the following:

1. The coefficients of several representative policy-sensitive equations in the MPS model have exhibited continuous movements over the past three decades. These movements in coefficients are a significant source of the total variation of important economic aggregates.

2. Even though continuous change in the model structure appears in all sample periods, the movements in key policy channels are larger in the 1980s. The sizable changes in the responses of the bond rate and a trade-weighted exchange rate to movements in the short-term interest rate in the 1980s support the focus in the body of the article on more recent structural changes.

3. The evidence does not support the simplifying assumptions that structural changes are wholly permanent or that the unpredictable changes are confined to a given historical interval. Unpredictable structural change occurs throughout the postwar sample. Indeed, in the 1980s, the proportion of variation due to unpredictable structural changes has increased for the effects of wealth on consumption and for the effects of the short-term interest rate on the bond rate and the exchange rate.

Why would one expect changes in economic structure to be continuous? Traditional econometric models assume that coefficients are fixed. In a macroeconomic context, fixed-coefficient models may be rationalized by the simplifying assumption that all economic agents are identical. However, it is more likely that the estimated coefficients of macroeconomic relationships are averages of the disparate responses of heterogeneous agents. For example, the aggregate response of consumption to income is an average over all households of the unique spending responses of each household. The aggregate response may be expected to vary over time because of changes in the cross-section distributions of household preferences or incomes.

The traditional fixed-coefficient model assumes also that the only sources of structural uncertainty are the intercepts of the estimated equations. (35) To continue with the example of aggregate consumption, only the subsistence level of consumption (represented by the intercept) is partially unpredictable under a standard fixed-coefficient specification. By contrast, the time-varying analysis allows the intercepts and the coefficients of explanatory variables, such as the slope multiplier of household income, to vary both predictably and unpredictably over time. (36) Predictable movements in the slope coefficients of explanatory variables are based on the estimated persistence of past coefficient changes and may include, as a special limiting case, the one-time slope-coefficient shifts specified in the body of this article. Unpredictable changes in slope coefficients, ruled out by assumption in fixed-coefficient analysis, provide a measure of the reliability of the estimated effects of explanatory variables.

Table A.1 [omitted] presents data-based evidence on the relative importance of continuous movements in the estimated structure of four key macroeconomic relationships in the MPS model. In each instance, the total variation of the dependent variable (household consumption, residential construction, the corporate bond rate, and a multilateral exchange rate) is attributed either to the fixed portion or to the continuously varying portion of the model structure. The fixed portion of the model structure is defined by the fixed coefficients originally estimated for the MPS model. Under the simplifying assumption that the explanatory variables are known, all fluctuations of the dependent variables described by the fixed portion of the model structure are predictable. The variation associated with the time-varying portion of the model structure is further disaggregated into variation due to the intercept and variation due to the slope coefficients. In contrast to the wholly predictable effects of the fixed portion of the model structure, the continuous movements in the intercept and slope coefficients are only partially predictable from past behavior. Under the convention that "predictable" denotes forecasts for one period ahead, table A.1 [omitted] presents an exhaustive allocation of the variation in a dependent variable to one of five factors: the fixed structure embodied in the MPS model, predictable movements in the intercept, unpredictable movements in the intercept, predictable movements in the slope coefficients, and unpredictable movements in the slope coefficients. (37)

The first three rows in table A.1 [omitted] examine allocations of the variation of household consumption in different periods. Over the full sample period (1960-88, line 1), the MPS fixed-structure rendering of the relationship among consumption, wealth, and income explains a sizable portion-96 percent--of the total variation in consumption, where much of the predictable variation in the level of consumption expenditures may be attributed to common trends in consumption and income. The remainder of the first row indicates that an additional 2 percent of the variation in aggregate consumption may be attributed to predictable changes in the time-varying influence of wealth and income, with no variation in consumption attributed to predictable movements in the intercept. Only 2 percent of the variation of total consumption is unpredictable, primarily because of unpredictable changes in the time-varying structural effects of wealth and income. As table A.1 indicates, the dominant source of slope variation in the consumption equation is associated with the coefficients of wealth. Thus, much of the structural variation of this equation includes a primary policy-transmission channel to consumption expenditures--the effect of movements in wealth induced by changes in interest rates.

The next two rows of the table contrast allocations of total variation of consumption in the 1980s and in earlier decades. The variation in consumption explained by the fixed MPS structure falls from 95 percent in the 1960s and 1970s to 82 percent in the 1980s. This decrease is partially offset by an increase in variation due to predictable changes in the time-varying coefficients of wealth and income. As the right-hand "total" columns show, total predictable variation in consumption fails to about 89 percent in the 1980s. The increase in net unpredictable variation of consumption is due mostly to substantially larger unpredictable movements in the time-varying effects of wealth. Because wealth is the dominant source of structural variability, the policy-transmission channel associated with interest rate effects on household wealth is more unpredictable in the 1980s. However, increased unpredictable variation in the response of consumption to wealth is not inconsistent with stable fixed-coefficient effects of wealth on consumption.

The next three rows in table A.1 [omitted] present allocations of variability for residential construction: The proportion of total variation explained by the MPS fixed structure increased from 67 percent in the pre-1980s sample to 78 percent in the 1980s. The residential construction equation used for the table incorporates no adjustments for the disintermediation effects of bank deposit rate ceilings in the 1960s and 1970s. Consequently, the sizable reduction in the 1980s of the intercept variation indicates that the effects of disintermediation in the earlier decades were largely captured by intercept shifts rather than by movements in the structural slope coefficients. Also, as shown by the allocations of variation associated with the intercept, quarter-to-quarter changes in the intercept were highly predictable in all sample periods, including the disintermediation effects embedded in intercept movements in the preperiod. 1980 Whereas time-varying movements in slope coefficients account for a portion of the variation in total residential construction, variations in the slope coefficients of the cost of capital were not the dominant source of structural change, as table A.1 [omitted] shows. Thus, the results of the time-varying specification appear to be similar to the results reported in text table 4, line 5 [omitted]; that line gives little evidence of time-varying changes in the effects of the cost of capital on residential construction if the disintermediation periods are represented by time-varying shifts in the intercept. (38)

As the next three rows of table A.1 show, the explanatory power of the MPS fixed structure in describing the behavior of corporate bond rates deteriorated in the 1980s. This deterioration has been accompanied by larger movements in the slope coefficients. Even if one accounts for the estimated persistence of past changes of the slope coefficients, the estimated response of long-term interest rates to movements in short-term interest rates apparently was not as predictable in the 1980s as in earlier decades. In fact, unpredictable changes in the slope-coefficient effects of short-term interest rates are virtually the sole source of uncertainty in the equation for corporate bond rates throughout the sample.

Finally, the last three rows of table A. I show the allocations of variation for the MPS exchange rate. As discussed in the body of this article, the dependent variable in this case is the exchange rate error produced by the MPS demand equations for foreign and domestic assets. This formulation removes the fixed-coefficient contribution of explanatory variables, including those interest rate variables embedded in the MPS specifications. To the extent that the spread between U.S. and foreign interest rates is correlated with excluded explanatory variables, the results in table A.1 cannot distinguish between structural change associated with the rate spread and structural change associated with excluded explanatory variables.

The overall predictability of the multilateral exchange rate was relatively stable from the 1970s to the 1980s, but the composition of predictable and unpredictable movements in the exchange rate underwent a sizable shift in the 1980s. The small fraction of predictable variation in the exchange rate that can be accounted for by the fixed-coefficient structure appears to have decreased in the 1980s. Partially offsetting this effect is an increase in variation associated with predictable changes in the slope coefficients of the interest rate spread, which is similar to the effect of the coefficient shift discussed in the article. In addition, however, the source of unpredictable variation in the exchange rate appears to have shifted in the 1980s from the intercept to the slope coefficients of the spread between U.S. and foreign interest rates. Thus, as with the long-term corporate bond rate, responses of the multilateral exchange rate to changes in the short-term interest rate appear to be less predictable in the 1980s.


Money Demand

Enzler, Jared, Lewis Johnson, and John Paulus. "Some Problems of Money Demand," Brookings Papers on Economic Activity, 1:1976, pp. 261-80.

Moore, George R., Richard D. Porter, and David H. Small. "Modeling the Disaggregated Demands for M2 and MI: The U.S. Experience in the 1980s," in Board of Governors of the Federal Reserve System, Financial Sectors in Open Economies: Empirical Analysis and Policy Issues. Washington: Board of Governors, 1990.

Porter, Richard D., Thomas D. Simpson, and Eileen Mauskopf, "Financial Innovation and the Monetary Aggregates," Brookings Papers on Economic Activity, 1:1979, pp. 213-29.

Exchange Rates

Meese, Richard. "Currency Fluctuations in the Post-Bretton Woods Era," Journal of Economic Perspectives, vol. 4 (Winter 1990), pp. 117-34.

Meese, Richard, and Kenneth Rogoff, "Empirical Exchange Rate Models of the Seventies: Do They Fit out of Sample?" Journal of International Economics, vol. 14 (February 1983), pp. 3-24. The "New" Credit Rationing

Jaffee, Dwight M., and Thomas Russell. "Imperfect Information, Uncertainty, and Credit Rationing," Quarterly Journal of Economics, vol. 90 (November 1976), pp. 651-66.

Stiglitz, Joseph E., and Andrew Weiss. "Credit Rationing in Markets with Imperfect Information," American Economic Review, vol. 71 (June 1981), pp. 393-410. Financial Variables and Real Activity

Fisher, Irving. "The Debt-Deflation Theory of Great Depressions," Econometrica, vol. I (October 1933), pp. 337-57.

Gurley, John G., and Edward S. Shaw. Money in a Theory of Finance. Washington: The Brookings Institution, 1960.

Keynes, John Maynard. The General Theory of Employment, Interest and Money, Chapter 12. New York: Harcourt, Brace, 1936.

Credit Rationing and the Transmission Channels

The discussion of the transmission channels of monetary policy focuses on the link between interest rates, asset prices, and real spending. The implicit assumption in the MPS model, as in much macroeconomic literature, has been that the financial system functions so efficiently that, except for the occasional disturbance due to institutional or regulatory factors--such as the episodes of disintermediation under Regulation Q--interactions between real and financial variables can be reduced and simplified to interactions between real variables and interest rates.

During the past decade, economists have influenced the debate about the transmission mechanism by their attention to the issue of information and, in particular, to the implications of imperfect information for the functioning of the financial system. An important proposition of this research is that using the interest rate to allocate credit may not be optimal when creditors have imperfect or incomplete information about debtors. This proposition seems most relevant in explaining investment decisions and suggests that for these decisions credit availability and cash flow may be more important, and the cost of capital may be less important, than the MPS model acknowledges. The alleged retrenchment of bank loans vis-a-vis the demand for business credit over the past year would fit this broadened view of the transmission mechanism. By emphasizing the role of financial variables other than interest rates, this approach has rejuvenated the interest in the connection between real activity and financial intermediation central to the work of John Gurley and Edward Shaw in the 1950s and important in the work of Irving Fisher and John Maynard Keynes in the 1930s (see the reading list at the end of the article).

Despite the theoretical plausibility of these new arguments for a more general view of the monetary transmission mechanism, and despite a series of studies using data from individual firms that have documented the importance of non-interest-rate credit rationing, empirical support for this view has yet to be found in the aggregate data. It is not clear whether this lack of support is due to problems of aggregation more severe than usual in applied work or to the relatively idiosyncratic occurrence of such rationing.


1. Other studies of the implications for monetary policy of the changing structure of the U.S. economy include the following: M.A. Akhtar and Ethan S. Harris, "Monetary Policy Influence on the Economy--An Empirical Analysis," Quarterly Review, Federal Reserve Bank of New York, vol. 11 (Winter 1986-87), pp. 19-34; Barry Bosworth, "Institutional Change and the Efficacy of Monetary Policy," Brookings Papers on Economic Activity, 1:1989, pp. 77-110; Benjamin M. Friedman, "Changing Effects of Monetary Policy on Real Economic Activity," Monetary Policy Issues in the 1990s (Federal Reserve Bank of Kansas City, August 30-September 1, 1989), pp. 55-111; and George A. Kahn, "The Changing Interest Sensitivity of the U.S. Economy," Economic Review, Federal Reserve Bank of Kansas City, vol. 74 (November 1989), pp. 13-34. In these studies, conclusions are mixed, with some sectors found to be less sensitive to interest rates over recent years and others to be more sensitive. The papers by Friedman and Kahn, the only ones that include an aggregate assessment, suggest that the net effect of these changes to sectoral interest-rate sensitivities is a reduced sensitivity of aggregate GNP to interest rates.

2. For a detailed description of the MPS model, see Flint Brayton and Eileen Mauskopf, "The Federal Reserve Board MPS quarterly econometric model of the US economy," Economic Modelling, vol. 3 (July 1985), pp. 170-292, and "Structure and Uses of the MPS Quarterly Econometric Model of the United States," Federal Reserve Bulletin, vol. 73 (February 1987), pp. 93-109. Probably the most important difference between the theory embedded in the MPS model and that espoused by one of the more popular schools of macroeconomic theory since the 1970s is in the modeling of expectations formation. In the MPS model, expected values of future variables are generally assumed to be based on past values of these variables. The alternative view--the rational expectations approach--argues that economic agents are motivated to use all available information in forming expectations, including their knowledge of the structure of the economy. Generally, the implementation of this approach sets expected values of future variables equal to the forecasts generated by the model in which the expectations appear.

3. Because equity and land values are the capitalized values of the income flow expected from the respective assets, the precise effect of a change in the federal funds rate on wealth depends on the level from which the interest rate is assumed to be reduced. For example, if the interest rate fell from 5.0 percent to 4.0 percent, the market value of wealth would rise 20 percent, given a sufficient length of time. If, instead, the interest rate were 9.0 percent, the percentage increase in wealth would be approximately 11 percent for the same 1 percentage point reduction. In tables 1 and 2, the federal funds rate averaged 7.7 percent in the base run so that the effective decline in the interest rate in tables 1 and 2 [omitted]is 13 percent.

4. Frank de Leeuw and Edward M. Gramlich, "The Channels of Monetary Policy: A Further Report on the Federal Reserve-MIT Econometric Model," Federal Reserve Bulletin, vol. 55 (June 1969), pp. 472-91.

5. In this simulation, monetary policy is characterized by a phased increase in the level of M2 of 1 1/2 percent, rather than a once-and-for-all decrease in the federal funds rate. The 1 1/2 percent increase in M2 is the average increase that results under the assumptions made in the top panel about prices and the federal funds rate.

6. Nevertheless, empirical evidence of a stable long-run relationship between M2 and nominal GNP supports a role for M2 as a nominal anchor for monetary policy. See Jeffrey J. Hallman, Richard D. Porter, and David H. Small, M2 per Unit of Potential GNP as an Anchor for the Price Level, Staff Studies 157 (Board of Governors of the Federal Reserve System, 1989).

7. Franco Modigliani and Robert J. Shiller, "Inflation, Rational Expectations and the Term Structure of Interest Rates," Economica, vol. 40 (February 1973), pp. 12-43. Of course, long-term interest rates do respond permanently to changes in the inflation rate to the extent that changes in the inflation rate are reflected in the commercial paper rate.

8. The higher the current short-term rate, the greater is the contribution of near-term coupon payments to the present value of the bond, and thus the shorter the duration of the bond. The hypothesis is that the duration of a bond will influence the manner in which past values of interest rates affect expected values. Specifically, for a bond of long duration, the behavior of interest rates over a long period of history is important in forming expectations about future rates; for a bond of shorter duration, the more recent pattern of interest rates is important for the formation of expectations.

9. Because the distributed lag weights on the commercial paper rate are a function of the level of recent commercial paper rates, an assumption must be made about the level of the paper rate in order to compute the distributed lag weights. We assumed that the commercial paper rate averaged 10 percent in both subperiods. The exact relation between the distributed lag weights and the level of the commercial paper rate appears, however, to have changed between the two subperiods.

10. The earnings-price ratio is commonly used as a measure of the real rate of return on equity. Because it is highly cyclical, it is probably not a good measure of the long-run expected return on equity. The dividend-price ratio is far less cyclical. Thus, the equation uses the product of the dividend-price ratio with an estimated inverse of the average payout ratio as a proxy for the long-run earnings-price ratio.

11. Freely estimated through 1973, the coefficient sum on the bond rate is not statistically different from 1.0; thus it is constrained to 1.0 in the version of this equation used in the MPS model. With this constraint, the coefficient sum on the inflation rate is estimated to be -0.89.

12. Franco Modigliani and Richard A. Cohn, "Inflation, Rational Valuation, and the Market," Financial Analysts Journal, vol. 35 (March/April 1979), pp. 24-44.

13. Over the estimation period 1963:2-1988:4, the coefficient sum on the bond rate is 0.74 and the sum on the inflation rate is 0. 15. Over the period ending in 1979:4, these sums are 0.95 and 0.28 respectively; over the 1980s, these sums are 1.08 and -0.14 respectively.

14. Attempts to improve the properties of the equation have been unsuccessful. For instance, when we try different proxies for expected inflation, the estimated long-run coefficient on expected inflation is statistically different from minus one and is often not statistically different from zero. If the long-run coefficient is constrained to minus one, the equation produces large errors. Although the possibility exists that our specification of the risk premium is incorrect, that possibility is unlikely to explain why the inflation coefficient, when freely estimated, is close to zero.

15. In another test, which permits coefficients on all explanatory variables to differ across the two subperiods, there is an even sharper decline in the coefficient on the contemporaneous value of the bond rate-from 0.78 in the early period to 0.42 in the later period.

16. To derive the exchange rate errors made by the model equations, the equations for asset demands are simulated given historical values for the current account, asset holdings (private and official), and interest rates. The simulated value of the exchange rate is that which is consistent with equality of the current account balance and the change in the net asset position.

17. This section presents a change in method. Rather than estimating the equations over the different subperiods and applying the standard F test to confirm the significance of any change, the prediction errors are analyzed to indicate the nature of behavioral shifts. We make this change because of the uncertainty about the reliability of the asset data, especially the breakdown of the net foreign asset position into U.S. holdings of foreign assets and foreign holdings of U.S. assets. When we simulate the two equations for asset demands, the errors tend to be offsetting so that the errors for the net position or, consequently, the errors in the exchange rate, are considerably smaller in percentage terms than those of the individual asset demand equations. Under these circumstances, applying the standard F tests to the two asset demand equations is unlikely to yield meaningful results.

18. See Federal Reserve Bulletin, various issues, table 1.54 (Mortgage Debt Outstanding). These figures include the principal balances of mortgage pools backing securities that are insured or guaranteed by the Government National Mortgage Association, the Federal Home Loan Mortgage Corporation, the Federal National Mortgage Association, and the Federal Housing Administration, as well as the principal balances of private pools.

19. Changes in the overall sensitivity of this sector to monetary policy also depend on changes in the behavior of the mortgage rate. In the MPS model, the mortgage rate is linked tightly to the corporate bond rate. As described in the preceding section, the corporate bond rate appears to have reflected changes in short-term interest rates more quickly in the 1980s than in earlier decades. The same is therefore true for the mortgage rate.

20. The effects of credit rationing on expenditures are captured by a second-degree distributed lag on a dummy-shift variable. The dummy-shift variable takes on a nonzero value during periods when credit appears to have been constrained, based on deposit flows at thrift institutions and on the institutional and regulatory structure in place. The periods of credit constraints are 1966:3-4, 1969:3-1970:3, and 1974:1-1975:1.

21. The intercept is permitted to differ in the two subperiods because the introduction of ARMs may have enlarged the pool of potential home buyers besides possibly altering the interest sensitivity of home buyers. The intercept would capture the former effect.

22. The estimated coefficients indicate that an increase of 1 percentage point in the mortgage rate directly reduces residential construction expenditures 16 percent in the early period and 9 percent in the later period, in each case within one year of the initial rise in the rate. The difference is not statistically significant. In a second test to detect change in the sensitivity of the desired housing stock to the cost of capital, the intercept is constrained to be unchanged over the entire period. The probability that the response to the cost of capital is unchanged over the entire period rises to 99 percent.

23. These calculations are an upper estimate of the importance of credit rationing for the overall interest sensitivity of housing expenditures. When the equation is estimated allowing only the cost-of-capital coefficient to differ over the two subperiods, the estimated coefficient on the cost of capital is less sensitive to the presence or absence of an explicit representation of disintermediation. However, because the introduction of ARMs may have led to a shift in the intercept (that is, to a larger pool of buyers), the specification that allows the intercept to change between the two subperiods is preferred.

24. Although the estimate of a 9 percent response is obtained both when the credit proxies are included and when they are excluded, no built-in constraint guaranteed this result. Because the equation is estimated over the entire sample period, all coefficient estimates could in practice be sensitive to the inclusion or exclusion of the credit proxies. For instance, eliminating the proxies for credit rationing might have changed the estimated income elasticity of housing (depending on the correlation between credit rationing and aggregate income) and this, in turn, might have affected the estimated coefficient on the cost of capital (depending on the correlation between income and the cost of capital) even over the later subperiod.

25. The focus here is on the subset of construction spending that excludes expenditures on petroleum and mining, which are modeled separately, and public utility investment other than telephone and telegraph investment, which is exogenous in the model.

26. The short- and intermediate-run elasticities depend on the estimated lag structure. These estimates are such that these elasticities tend to exceed their long-run values.

27. If one tests for structural shift before and after 1980, as we do for most cases, then the probability of no change in adjustment speed is higher. But this finding reflects the stability of the equation through 1985 and ignores the obvious problem of the fit since 1986.

28. For instance, the shorter lives for tax depreciation introduced by the Economic Recovery Tax Act in 1981 reduced the cost of capital. In response, construction activity increased. Unless the owner of the building lowered the rents, the demand for space would not increase with the supply of space. However, the consequent rise in vacancy rates and reduction in aggregate rent per building would not mean that the owner was then incurring a loss on the building: The change in the tax law implies that the rent on the building that is necessary to cover the costs would fall. The lower breakeven rent could be obtained by higher vacancy rates and unchanged rent per unit occupied or by unchanged vacancy rates and lower rent per unit occupied (or some combination of the two). However, the boom in construction subsequently set off a wave of rent reductions as each owner tried to find clients for the vacant office space. Ultimately, the combination of lower rents per unit occupied and higher vacancy rates reduced rents per building below break-even levels.

29. The more technical reader will want to distinguish between the elasticity of long-run investment with respect to the interest rate and the absolute change in long-run investment with respect to the interest rate--and, in particular, the effect of a shortening lifespan for equipment on both of these. In the long run, investment may be expressed as X(g+ [delta])/ r+8) where X is the level of output, g is the growth rate of real output, is the rate of depreciation of capital, and r is the real rate of interest. In this expression, we assume that the elasticity of substitution between capital and labor is unity, so that X/(r+ [delta]) is the optimal capital stock. The interest elasticity of long-run investment equals the interest elasticity of the optimal capital stock. This elasticity is negative but is less negative the shorter the lifespan of capital is. By contrast, the dollar response of long-run investment to a percentage point increase in the interest rate is negative, but shortening the lifespan of capital does not always reduce the size of this effect. The response depends on the relations among the growth rate of the economy, the real rate of interest, and the rate of depreciation. The intuition is that although a rise in the real rate of interest causes a smaller reduction in the stock of

30. Even if households are liquidity constrained, the effect on consumption of the incremental stock of adjustable-rate debt (measured as the average stock of adjustable-rate debt over the 1980s relative to that over the 1970s) is small. Under the extreme assumption that debt-servicing costs are met one for one by a reduction in consumption spending, the rise in debt-servicing costs for a 1.0 percentage point rise in interest rates is estimated to reduce consumption spending in the 1980s by 0.2 percentage point more than it did before 1980.

31. In the life-cycle model, the propensity to consume out of income and wealth depends on the rate of interest. If the relationship between the rate of interest and the propensity to consume out of wealth is assumed to be linear, then separate variables for property income (which is the product of the interest rate and the stock of wealth) and wealth belong in the equation.

32. As was hypothesized for nondurable consumption, one can argue that the effect of monetary policy on durables purchases may have changed because income has become more interest sensitive. To the extent that permanent income is more important than current income in explaining durables purchases, this effect is moderated.

33. The two versions of the model differ in the following ways: (1) The short-term interest rate affects the long-term rate more quickly in the post-1980s version; (2) the bond rate affects the return on equity more slowly in the post-1980s version; (3) preferences for assets denominated in specific currencies are more sensitive to interest rate differentials in the post-1980s version; the foreign interest rate is exogenized in both versions; (4) the sensitivity of residential construction expenditures to the cost of capital is smaller in the post-1980s version; and (5) nonresidential investment does not respond to the cost of capital in the post-1980s version.

34. One interpretation of a one-time shift in structural coefficients is that it is a special case of random walk coefficients in which all unpredictable changes in the coefficients are confined to a preselected historical interval. In general, the timing of the shift will be unknown to the modeler. A one-time-shift analysis will more reliably detect a permanent shift if the preselected location of the estimated break is close to the true break and there are sufficient observations before and after the true break.

35. An example of a fixed-coefficient equation is the simplified Keynesian consumption function

C(t) = a(t) + bY(t),

where consumption in period t, C(t), is a linear function of a time-varying intercept, a(t), and the product of a fixed slope coefficient, b, times the explanatory income variable, Y(t). By contrast, a time-varying coefficient formulation of the Keynesian consumption function is

C(t) = a(t) + b(t)Y(t).

where the slope coefficient, b(t), is now allowed to vary over time. In the time-varying specification, the slope and intercept include both fixed (a,b) and time-varying [u(t), v(t)] components

a(t) = a + u(t)

b(t) b + v(t)

whereas v(t) is set to zero in fixed-coefficient analysis. In fixed-coefficient models, the time-varying intercept contains a component that is predictable from past observed behavior and a component that is not. For example, a specification used widely in the MPS model is

u(t) = pu(t-1) + e(t),

where u(t- 1) denotes last period's error term, and e(t) is the unpredictable portion of the current time-varying intercept. In time-varying coefficient models, a similar decomposition into predictable and unpredictable elements is extended to the time-varying component of the slope coefficient, v(t).

36. The method of volatility allocation is based on the regression estimator developed by P.A.V.B. Swamy and P.A. Tinsley, "Linear Prediction and Estimation Methods for Regression Models with Stationary Stochastic Coefficients," Journal of Econometries, vol. 12 (1980), pp. 103-42.

37. The allocations of unpredictable volatility in table A.1 apply only to forecasts of economic activity over the near term (one quarter). Uncertainty about near-term forecasts is usually less than uncertainty about long-term forecasts because the predictable components of the slope and the intercept generally become less important as the forecast horizon lengthens. Diminishing predictability is a natural consequence of the decreasing dependence of today's outcome on outcomes in the more distant past. With certain exceptions, such as near-random-walk coefficients, the allocations to "predictable" variations in table A.1 become wholly unpredictable in lengthy forecast horizons.

Given the assumption that the explanatory variables of each model relationship are known, forecast errors are allocated only over unpredictable changes in the slope coefficients or over unpredictable movements in the intercept. These movements may be correlated because of common influences on the slope coefficients and on omitted variables captured by the moving intercept. In this case, one-half of the estimated co-movement is allocated to each. If the co-movement is negative and large, the net volatility contribution may be negative for either the time-varying slope coefficients or the intercept.

38. The disintermediation episodes in the 1960s and 1970s were triggered when market interest rates breached the deposit rate ceilings imposed by Regulation Q. However, the effects of disintermediation episodes are captured in the time-varying case by movements in the intercept, not by movements in the coefficients of the cost of capital. Thus, although the timing of disintermediation episodes may surely be attributed to positive spreads between market rates and the Regulation Q ceilings, the subsequent effects of disintermediations were not systematically related to the size of the rate spreads. Whether disintermediation is more appropriately modeled as changes in the slope coefficients of the cost of capital or as changes in the intercept remains an open issue, but either characterization is consistent with shifts in the conventional transmission channels of interest rates during the historical intervals of deposit disintermediation.

Eileen Mauskopf of the Board's Division of Research and Statistics prepared this article. The author acknowledges the research assistance of Sandra Cannon and the advice and assistance of Flint Brayton. Jeffrey Fuhrer and Peter Tinsley wrote the appendix.
COPYRIGHT 1990 Board of Governors of the Federal Reserve System
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1990, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:includes related information on credit rationing
Author:Mauskopf, Eileen
Publication:Federal Reserve Bulletin
Date:Dec 1, 1990
Previous Article:Record of policy actions of the Federal Open Market Committee.
Next Article:The current fiscal situation in state and local governments.

Related Articles
Record of policy actions of the Federal Open Market Committee.
Monetary policy in an era of change.
Statements to Congress.
Minutes of the Federal Open Market Committee Meeting of March 23, 1993.
Statement by Alan Greenspan, Chairman, Board of Governors of the Federal Reserve System, before the Committee on Banking, Housing, and Urban Affairs,...
Statement by Alan Greenspan, Chairman, Board of Governors of the Federal Reserve System, before the Committee on the Budget, U.S. House of...
Statement by Alan Greenspan.
Statement by Alan Greenspan, Chairman, Board of Governors of the Federal Reserve System, before the Commerce, Consumer, and Monetary Affairs...
New giant on wall street: the explosive growth of hedge funds is being felt from stock trading floors to the fed boardroom.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |