On the day of its release, the preliminary estimate of the Department of Commerce composite index of leading indicators (CIA) is widely reported in the popular and financial press. Although declines in the composite leading index are often regarded as a potential signal of the onset of a recession, evaluations of the ability of the CLI to predict turning points have been limited in most previous studies by the use of final, revised CLI data. However, the composite leading index is extensively revised after each preliminary estimate; not only are revisions made as more complete historical data become available for the components, but ex post, the statistical weights are updated and components are added or eliminated to improve leading performance. Forecasts constructed with an ex post, recomputed CLI may differ from real-time forecasts based on the contemporaneous, original construction CLI. In this chapter, we perform a completely ex ante, or real-time, evaluation of the ability of the CLI to predict turning points by using the original preliminary estimates and revisions as they became available in real time.

In section 14.1, we describe revisions in the CLI and our procedure for generating ex ante turning point probability forecasts from the CLI. The methodology is the Bayesian procedure described in Diebold and Rudebusch (1989), adapted to a newly constructed ex ante dataset. This new dataset, which has over 70,000 elements, contains every preliminary, provisionally revised, and final estimate of the CLI since the inception of the index in 1968.

From the point of view of a time-series analyst there are two questions of interest that concern the methodology dealing with the use of leading and coincident indicators. The first is the “surprising” longevity of the approach while competing methodologies such as large-scale econometric models or time-series representations have been subject to a fair amount of criticism over the years. This issue, we shall see, can be disposed of easily. It is the second question that is more substantial. Is the main contribution of the leading indicators approach to forecasting business cycle turning points one of convenience, or is it that the leading indicators capture an aspect of turning point prediction that econometric models or time-series representations miss?

Why is it not surprising that the methodology of leading indicators seems to have been subject to less controversy than the econometric models or time-series representations? Mainly because the latter two approaches impose explicit restrictions on observed phenomena and hence subject themselves to productive criticism and improvement. The approach of leading indicators, on the other hand, imposes few, if any, restrictions on the reality and is more robust. But it also generates less controversy and leads to fewer advances in understanding economic phenomena.

These comments notwithstanding, it is possible that the methodology of leading indicators captures an aspect of real phenomena that econometric models or time-series representations miss. This chapter deals with this question and shows under what conditions the approach has some special raison d'être.

Long-leading versus short-leading indicators

Most of the leading indicators that have been in use for many years have relatively short leads, averaging about six or eight months at business cycle peaks, when recessions begin, and two to four months at troughs, when recoveries start. Since there are often delays of a month or so in reporting the figures, and even longer delays in judging whether a turn in an indicator is significant, a recession or recovery may be well under way before it can be recognized. One way to deal with this problem is to distinguish indicators with exceptionally long leads from others.

The new long-leading index currently published by the Center for International Business Cycle Research takes a step in this direction. Using the revised list of fifteen leading indicators (Moore, 1989), we classified as long-leading those that had average leads of at least twelve months at peaks and six months at troughs during 1948–82. The four indicators that qualified as long-leading were bond prices, real money supply (M2), new building permits for housing, and a profit margin indicator, the ratio of prices to unit labor costs in manufacturing. A longleading index constructed from these series is shown in Figure 8.1, together with the short-leading index based on the other eleven series. Also shown is the Department of Commerce leading index as revised in March 1989, which contains two of the series in our long-leading group (money supply and housing permits), seven series in our short-leading group, and two other series.

It has long been recognized that there is something special about cyclical turning points and the predictions that are associated with these turns. It has been argued that the procedures for making quantitative forecasts, which are usually derived from minimum mean square error linear prediction theory, are not appropriate for making turning point predictions (Samuelson, 1976; Wecker, 1979). Since the process of detecting changes in regimes is different from making quantitative predictions, forecasting methods designed exclusively to recognize and predict turning points have been developed. These alternative methods, specifically designed for predicting turning points, include individual leading series, indexes of leading series, and rate of change methods.

Yet the problem remains: How do you determine whether a set of quantitative forecasts did or did not predict the onset or end of a recession? This chapter will explore this and related problems. The first section will review the criteria for identifying turning points when indicators are used. Then we will consider the difficulties that may be encountered if one has to determine whether quantitative forecasts predicted turns. The forecast record in the vicinity of turning points will be considered, and hypotheses to explain why the observed errors occurred will be advanced. Finally, there will be suggestions for improving forecasting performance.

Turning point predictions with indicators

Calling a turn

In analyzing the forecasting record of an indicator, it is necessary to develop rules for identifying turning points in the predictor.

This chapter is the result of attempts to find a way to extract as much information as possible about future economic activity from the twelve time-series commonly referred to as the leading indicators.

Leading indicators are data series that tend to lead business activity. Generally, it is believed that changes in certain economic variables precede, in a causal fashion, other economic variables. Because some of the leading series may produce false signals of future changes, it is also believed that an index composed of several of these leading series, chosen from a variety of economic processes, provides a better indication of future activity than any one series.

The most important of the single series used by business economists to forecast future economic activity is the so-called Index of Leading Indicators. This series is calculated by the Department of Commerce and published in the Survey of Current Business. The monthly announcements of the latest Index figure and the changes in the series composing it are major economic events. This chapter is an attempt both to test the forecasting ability of the Index and to create new indices that are more useful in forecasting peaks and troughs in the business cycle.

Section 11.1 is a review of some of the earlier attempts to use leading indicators to forecast economic activity. In section 11.21 present the theoretical basis for my own research and the results of my attempts to use the indicator series to forecast turning points in the business cycle. Section 11.3 presents my conclusions.

Correct prediction of the inflation rate and its turning points is an important problem for businesses and households alike. An early signal for a major turn in the inflation rate will allow economic agents to redo their economic calculations for the forthcoming environment. Because inflation rates are highly cyclical, Moore (1983a, 1983b) adapted the leading indicator approach, long associated with the National Bureau of Economic Research (NBER) studies of business cycles, to specifically forecast the inflation rate. Klein (1986) successfully extended this methodology of inflation forecasting to a number of major market-oriented economies. Moore (1986) reported that the composite inflation indicator has a better ex post record in forecasting next year's inflation rate than the consensus of economists has achieved. In Chapter 16, Roth evaluated five different leading indicators of inflation and found that composite indicators have a very impressive track record.

The main purpose of this chapter is to propose another predictor of inflation obtained by extracting information about future inflation from nominal interest rates of various maturities. For the sake of comparison, we will analyze these forecasts in the context of the existing leading indicator literature. Since the nominal interest rate, which is known at the beginning of a period, can be written as expected real interest rate plus expected rate of inflation, a reasonable estimate of the ex ante real rate will yield an equally reasonable estimate of the inflation component.

Many economic decision-makers seem to attach special importance to the phase of the business cycle – whether the economy is in an expansion or a recession – above and beyond the exact magnitude of the change in economic activity. The Gramm-Rudman-Hollings legislation, for example, contains a special provision suspending the budget deficit targets in the event of a recession. This concern about the stage of the business cycle is both a cause and an effect of the pioneering work by the National Bureau of Economic Research in identifying and dating cyclical turning points, the times when the economy has shifted from one phase to another. The Bureau's cycle chronology reaches back to 1854 on a monthly basis and even further on an annual basis.

The origin of this empirical approach predates the development of the National Income and Product or GNP Accounts and thus predates the development of macroeconometric models of GNP. While the designation of cyclical turning points derives from analyses of monthly data, the quarterly GNP accounts are the language of most macroeconomic models and forecasts.1 This presents a problem for those who ask, as many invariably do, how well cyclical turning points have been predicted. Most forecasts do not provide an explicit statement of when the economy will change from one phase to the next, and any attempt to assess their accuracy must first select the configuration of macroeconomic forecast data that can be taken as an indication that a cyclical turning point will occur.

The growing importance of the service industries justifies more attention being devoted to the development of indicators that reflect their part in economic fluctuations. In this chapter an attempt is made to apply to the service industries the indicator techniques that are well developed for analysis of the total economy. Leading, roughly coincident, and lagging indicators are identified for Australia's service industries. Composite indexes are constructed of five leading and four coincident indicators of the service industries. Both indexes are compared with the leading and coincident composite indexes that reflect the economic fluctuations in the total economy for Australia. The problems encountered in identifying service industry indicators are discussed.

One interesting finding is that the leading index for the service sector foreshadows equally as well as does our leading index for the total economy the pending changes in the pace and level of business activity. On the other hand, the coincident index for the service industries does not perform as consistently as does our coincident index for the total economy. However, both the leading and coincident indexes for the service sector are preliminary. Further work is being undertaken to improve their performance and also to identify more lagging indicators for the service sector. Nevertheless, it is believed that the results presented in this chapter help to clarify the role of the service industries in Australia's economic fluctuations.

Changes in commodity prices have long played an important indicative role in analyses of global economic conditions, principally because of their importance for developing countries. More than seventy countries derive at least 50 percent of their export earnings from nonfuel primary commodities; another twenty derive the majority of their export earnings from fuels (see IMF, 1988, pp. 104–5). Changes in the terms of trade for these countries typically arise largely from changes in world commodity prices. Recently, however, attention has also been drawn to the importance of changes in commodity prices as indicators of changes in inflationary conditions affecting industrial countries. For example, the World Economic Outlook recently began to include an analysis comparing percentage changes in an index of forty primary commodity prices with the aggregate inflation rate of the seven largest industrial countries (see IMF, 1988, p. 11). The task of this chapter is to examine the usefulness of commodity prices as a leading indicator of inflation in the large industrial countries as a group.

An early exponent of focusing on commodity prices in this context was Robert Hall. In his 1982 book, Hall argued in favor of basing U.S. monetary policy on a commodity standard, with the commodities chosen on the basis of the closeness of their historical fit against the cost of living. Bosworth and Lawrence (1982) also emphasized the role of commodity prices as a contributor to the rise in inflationary pressures during the 1970s.

Early detection or even timely recognition of business cycle turning points has always been a major concern of policy makers, businesses, and investors. Clearly, early recognition would allow the government policy maker to trigger countercyclical policy measures, businesses to change their own sales or investment strategy, and investors to reallocate assets among alternative investments to optimize their return. The typical way of monitoring and forecasting cyclical turning points is to use leading indicators. Unfortunately, no leading indicator is 100 percent perfect, which means it is sometimes difficult to tell whether or not the leading indicator signal is real. Over the years, numerous systems have been developed to screen out false signals. When these systems were put to the real-life test of forecasting turning points, some of these systems have worked well while others have not. Nearly all the methods for screening turning point signals have been ad hoc creations that may or may not have credibility with other users. However, there is one method, proposed by Salih Neftci of City University of New York (CUNY), that adds a new dimension to screening out false signals. This method is based on economic theory and statistical methods.

Neftci has proposed a method that uses sequential analysis to calculate the probability of a cyclical turning point. This method is based on a theoretical and empirical claim that the onset of a recession is marked by a pronounced decline in aggregate economic activity.

The leading indicator approach

The leading indicator approach to economic and business forecasting is based on the view that market-oriented economies experience business cycles within which repetitive sequences occur and that these sequences underlie the generation of the business cycle itself. Wesley Mitchell (1927), one of the founders of the National Bureau of Economic Research (NBER), first established a workable definition of business cycles, and Burns and Mitchell (1946) rephrased it as follows:

The leading economic indicator (LEI) approach then is to find the repetitive sequences, to explain them, and to use them to identify and to forecast emerging stages of the current business cycle.

The approach differs from the usual econometric model, which does not differentiate business cycles from other economic fluctuations, except perhaps seasonal variation.

The immediate origin of this book was a conference held at the State University of New York at Albany in May 1987 at which a number of well-known exponents of leading indicators research were invited to offer their views on this subject. During the conference the need was felt for a volume that would systematically explain and evaluate the old and the new emerging techniques dealing with leading economic indicators. Thus, the volume is not the proceedings of the conference, but rather a collection of mostly previously unpublished articles broadly representing current research in this field. A number of authors were commissioned to write chapters on specific topics, and the two editors undertook to provide an appropriate framework. Many of the economists whose work has been central to the development of leading indicators report new research, review progress in specific areas, and discuss directions for further work. We hope that the book will prove useful to university economists and students interested in business cycles and forecasting, as well as to those in business firms, government agencies, and international organizations who wish to keep abreast of new developments in this field and adapt them for practical use.

The editors are grateful to the authors for their contributions, their involvement in revising their own chapters, and their patience during the course of the production of the book.

The Gramm-Rudman-Hollings (GRH) law passed by the Congress in December 1985 establishes a process whereby the Federal budget deficits are to be gradually phased out by the fiscal year (FY) 1991. A series of targeted ceilings on the unified budget deficit is instituted, beginning with $172 billion for FY 1986 and $144 billion for FY 1987 and proceeding by decrements of $36 billion per year to zero in FY 1991. The planned reductions are to be achieved by spending and tax measures agreed upon by the legislative and executive branches of the U.S. government. However, if an agreement is not reached, the target for any FY is to be achieved through an automatic across-the-board spending cut in all eligible defense and nondefense categories. Early in 1986 a lower court ruled that the sequestering provision of GRH is unconstitutional; this ruling was confirmed by the Supreme Court on July 7, 1986. This does not pertain to the issues discussed in this paper.

Section 254 of the GRH law provides for “Special Procedures in the Event of a Recession.” It states that the Congressional Budget Office (CBO) Director shall notify the Congress at any time (a) if the CBO or the Office of Management and Budget (OMB) “as determined that real economic growth is projected or estimated to be less than zero with respect to each of any two consecutive quarters” within a period of six successive quarters starting with the one preceding such notification, or (b) “if the Department of Commerce preliminary report of actual real economic growth (or any subsequent revision thereof) indicates that the rate of real economic growth for each of the most recent reported quarter and the immediately preceding quarter is less than one percent.”