# Indexes of coincident and leading economic indicators.

Indexes of Coincident and Leading Economic Indicators(1)

For 50 years, economists in business and government have used the system of leading economic indicators to gauge the future course of economic activity. The system of leading, coincident, and lagging economic indicators originally was developed by Arthur F. Burns, Wesley C. Mitchell, and their colleagues at the NBER and is currently maintained by the U.S. Department of Commerce (DOC). Some 32 countries throughout the world now have a system of indicators that they use. The indexes of coincident and leading economic indicators themselves--weighted averages of key coincident and leading time series--play a central role in contemporary uses of this system. The coincident index measures the current state of the economy. The leading index often is interpreted as giving advance information about the future direction of the economy, particularly whether toward an expansion or a recession.

In recent work, we have taken a new look at the construction and interpretation of the indexes of coincident and leading economic indicators. The methods used to construct these indexes have remained largely unchanged for the last 30 years. We have exploited recent developments in time-series econometrics to improve the performance of the coincident and leading economic indexes constructed using traditional techniques. This work has resulted in the development of three experimental indexes: an index of coincident economic indicators (CEI), an index of leading economic indicators (LEI), and a new series that we call a "recession index" (RI). These three indexes, their construction, and their interpretation are described in this Research Summary.

The Index of Coincident Economic Indicators

In constructing an index of leading indicators, the first step is to define what this index leads. The index of coincident indicators currently maintained by the DOC is a weighted average of four broad measures of economic activity: industrial production, real personal income less transfer payments, real manufacturing and (1)This report draws on research reported in J.H. Stock and M.W. Watson, "New Indexes of Coincident and Leading Economic Indicators," presented at the NBER Macroeconomics Conference, 1989. This work was funded in part by the NBER. The results of this work are still experimental and do not constitute an official new set of NBER indexes. trade sales, and the number of nonagricultural employees. While each of the series exhibits its own idiosyncratic movements (which include errors of measurement), the common movement among the series may arise from general swings in economic activity, that is, from the business cycle. Thus averaging these series provides one way to eliminate the idiosyncratic movements and obtain a better estimate of swings in overall activity.

But how can this averaging best be done? The traditional NBER/DOC approach is to take a weighted average of contemporaneous growth rates of the coincident series, in which the weights depend on the standard deviations of the series. Although we chose to construct the weights somewhat differently--using an explicit statistical model--the net result is very similar to the DOC coincident index. (Our weights are from an estimated "dynamic factor model," in which the unobserved state of the economy is the sole source of comovements among the coincident variables.)(2) The major difference between the variables in our experimental index and the DOC index is that we use employee-hours rather than the number of employees. (2)In theory the traditional method and the "dynamic factor model" approach could have produced quite different indexes. The fact that the indexes are so similar can be interpreted as providing a formal statistical rationalization for the traditional procedure. The application of dynamic factor models to macroeconomic time-series variables was developed by T.J. Sargent and C.A. Sims, "Business Cycle Modeling without Pretending to Have Too Much A Priori Economic Theory," in C.A. Sims et al., New Methods in Business Cycle Research, Minneapolis: Federal Reserve Bank of Minneapolis, 1977. For details concerning the construction of the coincident indicator model, see J.H. Stock and M.A. Watson, "A Probability Model of the Coincident Economic Indicators, "in G.H. Moore and K. Lahiri, eds., The Leading Indicators: New Approaches and Forecasting Records, New York: Cambridge University Press, forthcoming.

The DOC coincident index and our experimental coincident index are plotted in Figure 1; both are scaled to equal 100 in 1967. The vertical bars in Figure 1 denote official NBER-dated peaks and troughs. The major difference between the experimental index and the DOC index is the slightly higher trend growth in the DOC index. The correlation between the monthly growth rates of the two series is high (the correlation coefficient is .95). Moreover, the timing of peaks and troughs in the two indexes is the same.

The Indexes of Leading Economic Indicators

The existing index of leading indicators serves two distinct purposes: to forecast the growth of the economy over the next several months, and to provide an early signal of an upcoming recession or expansion. Our experimental indexes separate these two functions: the experimental LEI is a forecast of the growth of the overall economy (as measured by the CEI) over the next six months, while the RI reports a probability of the economy being in a recession in six months.

We use seven leading series, selected from an original list of over 280 series, to construct the experimental LEI. Traditionally series for the leading index have been chosen based on their historical ability to lead some measure of overall activity, such as the coincident index. For our experimental LEI, we used this "bivariate" approach to screen possible series but relied on a "multivariate" criterion in developing the final list. This criterion identified variables that have information not contained in the other time series already in the experimental LEI but that have been useful historically for forecasting overall activity six months hence.

Of the seven variables in the experimental LEI, two are in the current DOC index: manufacturers' unfilled orders (durable goods industries) and new private housing authorizations.(3) Of the remaining five variables, three are based on interest rates: the spread between six-month commercial paper and six-month U.S. Treasury bills; the spread between ten-years Treasury bonds and one-year Treasury bonds; and the change in the ten-year Treasury bond rate. The final variables are part-time work in nonagricultural industries because of slack work and a trade-weighted index of exchange rates between the United States and the United Kingdom, West Germany, France, Italy, and Japan.

The experimental LEI (the forecast of the growth in the experimental CEI over the next six months, at annual rates, based on these seven variables) is plotted in Figure 2. Also plotted in Figure 2 is the actual six-month growth of the CEI. Like any forecast, the LEI is an imperfect map of future economic events. By comparing the two series, one can get a sense of when the experimental

LEI would have succeeded and when it would

" (3)The DOC revised its leading and lagging indexes in March 1989, for data starting January 1989; the coincident index was not changed. These remarks refer to the most recent revision. have failed. In the summer of 1979, for example, the experimental LEI became negative, indicating negative growth in the CEI over the next six months; in fact this is what occurred. In contrast, in early 1982 the experimental LEI hovered near zero, when in fact the economy continued to suffer a decline.

Interest rates play an important role in the LEI: an inverted Treasury bond yield curve and a high spread between short-term commercial paper and Treasury bills of a matched maturity are statistically important precursors of declines in overall economic activity. Interestingly, the statistical selection procedures that led to these seven series indicated that some traditional leading variables--in particular, the money supply (M2) and the growth of stock prices--have little additional forecasting value, once the information in the seven series already in the experimental LEI are taken into account.

The Recession Index

An important objective of this research has been to develop a new index that provides a direct assessment of whether the economy will slip into a recession. The Recession Index estimates the probability that the economy will be in a recession six months hence. This probability is calculated using the time series comprising the experimental CEI and LEI.

Two series that measure whether the economy is or will be in a recession are plotted in Figure 3. Figure 3(a) represents a series that answer the question: is the economy currently in a recession? That is, this series is the probability that the economy is in a recession in a given month, using data available through the end of between zero and one: a value of near one indicates that it is highly likely that the economy is, at that date, in a recession.

The series in Figure 3(b) answers a more difficult question: will the economy be in a recession six months hence? This is the experimental RI. Not surprisingly, the probabilities in Figure 3(b) are not as sharp as those in Figure 3(a). Still, based on historical data, the RI would have "predicted" each of the four recessions since 1960, although it incorrectly "predicted" one recession (in 1967) that did not occur.

Summary

These experimental indexes have been developed by closely examining historical patterns using the tools of modern econometrics. The emphasis in developing the LEI and the RI has been to exploit information in multiple time series, rather than to focus on the bivariate relationship between a given time series and the business cycle, one series at a time. In principle, this approach offers the possibility of substantial improvements in the prediction of recessions and expansions. By tracking the future performance of these indexes, we will be able to determine whether this possibility is realized.

For 50 years, economists in business and government have used the system of leading economic indicators to gauge the future course of economic activity. The system of leading, coincident, and lagging economic indicators originally was developed by Arthur F. Burns, Wesley C. Mitchell, and their colleagues at the NBER and is currently maintained by the U.S. Department of Commerce (DOC). Some 32 countries throughout the world now have a system of indicators that they use. The indexes of coincident and leading economic indicators themselves--weighted averages of key coincident and leading time series--play a central role in contemporary uses of this system. The coincident index measures the current state of the economy. The leading index often is interpreted as giving advance information about the future direction of the economy, particularly whether toward an expansion or a recession.

In recent work, we have taken a new look at the construction and interpretation of the indexes of coincident and leading economic indicators. The methods used to construct these indexes have remained largely unchanged for the last 30 years. We have exploited recent developments in time-series econometrics to improve the performance of the coincident and leading economic indexes constructed using traditional techniques. This work has resulted in the development of three experimental indexes: an index of coincident economic indicators (CEI), an index of leading economic indicators (LEI), and a new series that we call a "recession index" (RI). These three indexes, their construction, and their interpretation are described in this Research Summary.

The Index of Coincident Economic Indicators

In constructing an index of leading indicators, the first step is to define what this index leads. The index of coincident indicators currently maintained by the DOC is a weighted average of four broad measures of economic activity: industrial production, real personal income less transfer payments, real manufacturing and (1)This report draws on research reported in J.H. Stock and M.W. Watson, "New Indexes of Coincident and Leading Economic Indicators," presented at the NBER Macroeconomics Conference, 1989. This work was funded in part by the NBER. The results of this work are still experimental and do not constitute an official new set of NBER indexes. trade sales, and the number of nonagricultural employees. While each of the series exhibits its own idiosyncratic movements (which include errors of measurement), the common movement among the series may arise from general swings in economic activity, that is, from the business cycle. Thus averaging these series provides one way to eliminate the idiosyncratic movements and obtain a better estimate of swings in overall activity.

But how can this averaging best be done? The traditional NBER/DOC approach is to take a weighted average of contemporaneous growth rates of the coincident series, in which the weights depend on the standard deviations of the series. Although we chose to construct the weights somewhat differently--using an explicit statistical model--the net result is very similar to the DOC coincident index. (Our weights are from an estimated "dynamic factor model," in which the unobserved state of the economy is the sole source of comovements among the coincident variables.)(2) The major difference between the variables in our experimental index and the DOC index is that we use employee-hours rather than the number of employees. (2)In theory the traditional method and the "dynamic factor model" approach could have produced quite different indexes. The fact that the indexes are so similar can be interpreted as providing a formal statistical rationalization for the traditional procedure. The application of dynamic factor models to macroeconomic time-series variables was developed by T.J. Sargent and C.A. Sims, "Business Cycle Modeling without Pretending to Have Too Much A Priori Economic Theory," in C.A. Sims et al., New Methods in Business Cycle Research, Minneapolis: Federal Reserve Bank of Minneapolis, 1977. For details concerning the construction of the coincident indicator model, see J.H. Stock and M.A. Watson, "A Probability Model of the Coincident Economic Indicators, "in G.H. Moore and K. Lahiri, eds., The Leading Indicators: New Approaches and Forecasting Records, New York: Cambridge University Press, forthcoming.

The DOC coincident index and our experimental coincident index are plotted in Figure 1; both are scaled to equal 100 in 1967. The vertical bars in Figure 1 denote official NBER-dated peaks and troughs. The major difference between the experimental index and the DOC index is the slightly higher trend growth in the DOC index. The correlation between the monthly growth rates of the two series is high (the correlation coefficient is .95). Moreover, the timing of peaks and troughs in the two indexes is the same.

The Indexes of Leading Economic Indicators

The existing index of leading indicators serves two distinct purposes: to forecast the growth of the economy over the next several months, and to provide an early signal of an upcoming recession or expansion. Our experimental indexes separate these two functions: the experimental LEI is a forecast of the growth of the overall economy (as measured by the CEI) over the next six months, while the RI reports a probability of the economy being in a recession in six months.

We use seven leading series, selected from an original list of over 280 series, to construct the experimental LEI. Traditionally series for the leading index have been chosen based on their historical ability to lead some measure of overall activity, such as the coincident index. For our experimental LEI, we used this "bivariate" approach to screen possible series but relied on a "multivariate" criterion in developing the final list. This criterion identified variables that have information not contained in the other time series already in the experimental LEI but that have been useful historically for forecasting overall activity six months hence.

Of the seven variables in the experimental LEI, two are in the current DOC index: manufacturers' unfilled orders (durable goods industries) and new private housing authorizations.(3) Of the remaining five variables, three are based on interest rates: the spread between six-month commercial paper and six-month U.S. Treasury bills; the spread between ten-years Treasury bonds and one-year Treasury bonds; and the change in the ten-year Treasury bond rate. The final variables are part-time work in nonagricultural industries because of slack work and a trade-weighted index of exchange rates between the United States and the United Kingdom, West Germany, France, Italy, and Japan.

The experimental LEI (the forecast of the growth in the experimental CEI over the next six months, at annual rates, based on these seven variables) is plotted in Figure 2. Also plotted in Figure 2 is the actual six-month growth of the CEI. Like any forecast, the LEI is an imperfect map of future economic events. By comparing the two series, one can get a sense of when the experimental

LEI would have succeeded and when it would

" (3)The DOC revised its leading and lagging indexes in March 1989, for data starting January 1989; the coincident index was not changed. These remarks refer to the most recent revision. have failed. In the summer of 1979, for example, the experimental LEI became negative, indicating negative growth in the CEI over the next six months; in fact this is what occurred. In contrast, in early 1982 the experimental LEI hovered near zero, when in fact the economy continued to suffer a decline.

Interest rates play an important role in the LEI: an inverted Treasury bond yield curve and a high spread between short-term commercial paper and Treasury bills of a matched maturity are statistically important precursors of declines in overall economic activity. Interestingly, the statistical selection procedures that led to these seven series indicated that some traditional leading variables--in particular, the money supply (M2) and the growth of stock prices--have little additional forecasting value, once the information in the seven series already in the experimental LEI are taken into account.

The Recession Index

An important objective of this research has been to develop a new index that provides a direct assessment of whether the economy will slip into a recession. The Recession Index estimates the probability that the economy will be in a recession six months hence. This probability is calculated using the time series comprising the experimental CEI and LEI.

Two series that measure whether the economy is or will be in a recession are plotted in Figure 3. Figure 3(a) represents a series that answer the question: is the economy currently in a recession? That is, this series is the probability that the economy is in a recession in a given month, using data available through the end of between zero and one: a value of near one indicates that it is highly likely that the economy is, at that date, in a recession.

The series in Figure 3(b) answers a more difficult question: will the economy be in a recession six months hence? This is the experimental RI. Not surprisingly, the probabilities in Figure 3(b) are not as sharp as those in Figure 3(a). Still, based on historical data, the RI would have "predicted" each of the four recessions since 1960, although it incorrectly "predicted" one recession (in 1967) that did not occur.

Summary

These experimental indexes have been developed by closely examining historical patterns using the tools of modern econometrics. The emphasis in developing the LEI and the RI has been to exploit information in multiple time series, rather than to focus on the bivariate relationship between a given time series and the business cycle, one series at a time. In principle, this approach offers the possibility of substantial improvements in the prediction of recessions and expansions. By tracking the future performance of these indexes, we will be able to determine whether this possibility is realized.

Printer friendly Cite/link Email Feedback | |

Author: | Stock, James H.; Watson, Mark W. |
---|---|

Publication: | NBER Reporter |

Date: | Mar 22, 1989 |

Words: | 1629 |

Previous Article: | Business cycles. |

Next Article: | Asset-price expectations. |

Topics: |