Abstract
This Selected Issues paper and Statistical Appendix compares two alternative time series approaches to analyzing Switzerland’s recent business cycle experience: first, the traditional “smooth-trend-plus-cycle approach,” which envisages observed output growth as fluctuating around a relatively smooth potential output growth path; and, second, the more recently developed “regime change approach,” which views business cycles as shifts between “high-growth” states (expansions) and “slow-growth” states (recessions) of the economy. The paper also examines Switzerland’s monetary policy framework, and describes the challenges to the Swiss tax system.
I. Alternative Models of the Swiss Business Cycle1
A. Introduction and Summary
1. At the beginning of the 1990s, the real GDP growth rate of the Swiss economy slowed sharply below its average growth path (Figure I-1). The following period of output stagnation (1991-96) was prolonged, and unemployment rose to postwar records. The stagnation period also proved a taxing experience for business cycle forecasters and analysts: official and private forecasts of real GDP growth almost persistently overpredicted the future GDP growth rate of the economy; and assessments of the cyclical position of the economy based on the traditional decomposition of real GDP into trend and cyclical output components produced a wide range of estimates of the economy’s output gap.
Switzerland: Actual and Projected Real GDP Growth, 1987-97
(In percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Sources: IMF, World Economic Outlook database; and OECD Economic Outlook (various Issues).Switzerland: Actual and Projected Real GDP Growth, 1987-97
(In percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Sources: IMF, World Economic Outlook database; and OECD Economic Outlook (various Issues).Switzerland: Actual and Projected Real GDP Growth, 1987-97
(In percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Sources: IMF, World Economic Outlook database; and OECD Economic Outlook (various Issues).2. Against this background, this chapter compares two alternative time series approaches to analyzing Switzerland’s recent business cycle experience: first, the traditional “smooth-trend-plus-cycle approach,” which envisages observed output growth as fluctuating around a relatively smooth potential output growth path; and, second, the more recently developed “regime change approach,” which views business cycles as shifts between “high-growth” states (expansions) and “slow-growth” states (recessions) of the economy.2 While the smooth-trend-plus-cycle approach seeks to separate trend and cycle components and views recessions and expansions as mirror images of each other, the regime change approach assumes that trend and cycle are fundamentally intertwined phenomena and allows for asymmetric durations of expansions and recessions.
3. The empirical analysis reported in this chapter suggests that a regime change model fits the stylized facts of Switzerland’s recent business cycle experience better than the traditional trend-cycle model:
Consistent with the predictions of the regime change approach, Switzerland’s recent business cycle experience has been characterized by abrupt and persistent shifts in real GDP growth rates. Moreover, the frequency distribution for observed Swiss real GDP growth rates since the mid-1970s appears to have a distinct bimodal shape, with observed GDP growth rates clustering around “low-growth” and “high-growth” states.
The regime change model can account for the dismal record of macroeconomic forecasts in tracking future Swiss real GDP growth. A highly persistent pattern of forecast errors would be consistent with forecasters basing their predictions on a smooth-trend-plus-cycle model, which assumes that GDP growth has a pronounced tendency to revert back to potential growth, while output growth in fact alternates between persistent regimes of high and slow growth.
4. A regime change model of the Swiss business cycle may also provide a useful perspective for interpreting several features of Switzerland’s policy behavior and framework:
While a regime change model view of the business cycle supports the traditional medium-term orientation of Switzerland’s policy framework, it also would provide a rationale for the widespread view that particularly large unfavorable exchange rate or aggregate demand shocks (e.g., brought about by a collapse in consumer confidence) could have a severe and lasting impact on the average performance of the economy, as these type of shocks could shift the economy to a protracted slow-growth state. In fact, descriptions of Switzerland’s monetary policy framework often emphasize that while the SNB will aim at meeting a medium-term target for money growth, the central bank preserves the option of discretionary deviations from its medium-term money target in case of particularly large unfavorable shocks.3 By contrast, within the confines of the smooth-trend-plus-cycle approach, most adverse shocks should only be reflected in transitory dips in the economy’s output gap, and it is far less clear why these shocks should trigger discretionary policy responses.
A regime change view of the business cycle appears also to be implicit in the recent discussions of a desirable fiscal policy rule consistent with budget balance over the medium term (at the Confederation level).4 In particular, the authorities’ specific proposals for such a rule have been couched in terms of an (implicit) three-state regime change model, where the fiscal policy response to real GDP growth variations would differ depending on whether the economy is projected to be in a “recession state” (real GDP growth < 0.5 percent), in a “normal state” (real GDP growth ≥ 0.5 percent but smaller than 1.8 percent), or in a “boom state” (real GDP growth ≥ 1.8 percent).5
5. As regards predicting medium-term developments in output growth (and, as a likely consequence, the medium-term inflation rate), regime change models of the business cycle would suggest that forecasts are fraught with considerably more uncertainty than indicated by the smooth-trend-plus-cycle approach. The latter approach essentially implies that medium-term output forecasts can be based on the rule-of-thumb that future GDP growth is equal to the potential output growth rate plus an adjustment needed to close the output gap over the medium term. Under a regime change model, the durations of expansions and recessions can be asymmetric, and the exact timing of shifts between high- and slow-growth states appears to be difficult to predict with any confidence.
6. The chapter concludes that the economic forces that underpin the cyclical shifts between high- and low-growth regimes are at this point not well understood. Some researchers have conjectured that the persistence of cyclical expansions and recessions captured by regime change models may in part reflect “thick market” effects, i.e., economic activity is more efficient when it occurs in a concentrated fashion over time.6 In the particular case of Switzerland, several structural characteristics of the economy including a highly elastic labor supply (related to worker migration flows), a large construction sector with long gestation periods, an elastic supply of loanable funds owing to a large and sophisticated financial sector, and the pronounced procyclical behavior of fiscal policy may all have contributed to the observed periods of persistently slow and fast GDP growth.
7. The remainder of this chapter is organized as follows: Section B reviews recent experience with forecasting and assessing the cyclical position of the Swiss economy. Section C outlines the statistical properties of the smooth-trend-plus-cycle and the regime change approaches to modeling business cycle fluctuations. Section D presents empirical evidence. Section E discusses recent theoretical ideas and economic forces that could underpin the workings of regime change models. And Section F concludes with some implications of the regime change model approach for forecasting and policy analysis.
B. Recent Business Cycle Fluctuations
8. Switzerland’s recent business cycle experience has been characterized by abrupt and persistent shifts in real GDP growth. In particular, during the second half of the 1980s, real GDP growth expanded at a brisk rate (Figure I-1). This economic boom came to a sudden end in 1991, and was followed by a protracted six-year stagnation period in output. In 1997, the economy finally shifted back to a new phase of markedly stronger economic activity.
9. Macroeconomic forecasts of real GDP growth developments during this period were almost consistently off the mark, either underpredicting (during the boom period 1987-90) or overpredicting (during the stagnation period 1991-96) actual GDP growth (see Figure I-1).7 The highly persistent pattern of forecast errors most likely reflects an assumption on the part of forecasters that real GDP growth would revert quickly back to the average growth path of the economy. The persistent sequence of forecast errors in the case of recent Swiss real GDP growth notwithstanding, systematic forecast errors around business cycle turning points appear to be a perennial feature of output growth forecasts. For example, in an assessment of the Fund’s WEO forecasts, Artis (1997, pp. 22) concludes:
“Systematic turning point error taking the form of an initial underestimate or an overestimate of output growth, followed by persistence in the same error… is uncomfortably pervasive in the data.”
10. Assessments of the cyclical position of the Swiss economy based on estimates of potential output growth and the cyclical output gap were recently also subject to uncommonly large margins of uncertainty (Figure I-2). Business cycle analysis based on the smooth-trend-plus-cycle model would suggest that a slowdown in actual GDP growth signals usually a transitory downturn of economic activity. However, postulating that potential output growth in Switzerland during the stagnation period 1991-96 amounted to 2 percent—in conformity with the experience of the second half of the 1980s and the Swiss National Bank’s (SNB) medium-term base money target path for the five-year periods 1989-94 and 1994-99—would yield an implausibly large negative output gap of about 9 percent in 1998. On the other hand, allowing potential output growth to adjust flexibly to the actual GDP growth experience—as done by the Federal Finance Administration (FFA) in its Hodrick-Prescott filter-based estimates of the output gap used for the calculation of the structural budget balance—yields an output gap that was already somewhat positive in 1996. Finally, the staffs own assessment of potential GDP growth and output gap estimates—which are based on a pragmatic approach that allows some adjustment of potential to actual GDP growth during the 1990s—suggests that the output gap was about 1½ percent in 1998.8
Switzerland: Potential Output Growth and Output Gaps, 1987-98
(In percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Sources: IMF, World Economic Outlook database; and staff estimates.1/ Staff estimates.2/ 2 percent potential output growth; based on SNB’s monetary framework.3/ Hodrick-Prescott filter estimates of potential output growth and output gap.Switzerland: Potential Output Growth and Output Gaps, 1987-98
(In percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Sources: IMF, World Economic Outlook database; and staff estimates.1/ Staff estimates.2/ 2 percent potential output growth; based on SNB’s monetary framework.3/ Hodrick-Prescott filter estimates of potential output growth and output gap.Switzerland: Potential Output Growth and Output Gaps, 1987-98
(In percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Sources: IMF, World Economic Outlook database; and staff estimates.1/ Staff estimates.2/ 2 percent potential output growth; based on SNB’s monetary framework.3/ Hodrick-Prescott filter estimates of potential output growth and output gap.C. Statistical Models
Two specific versions of this general statistical model can now be used to contrast the traditional smooth-trend-plus-cycle (STPC) and the regime change (RC) approaches to analyzing business cycle fluctuations.
Smooth-trend-plus-cycle (STPC) approach
where Δyp(t) (=μ(t)) is potential output growth and ΔGAP(t) is the change in the cyclical output gap. Within the STPC approach, the key challenge is to identify the shocks that drive the unobserved potential and cyclical gap components, respectively. The literature has proposed a considerable number of procedures to accomplish this objective (see, e.g., Harvey (1989)). Depending on the identifying assumptions underlying the decomposition and the characteristics of the output series, the estimates of potential output growth resulting from different procedures can, however, vary substantially, as illustrated by the Swiss experience summarized in Figure I-2.
Regime change (RC) approach
13. By contrast to the STPC approach, regime change models of the business cycle allow the underlying mean output growth rate to shift between “slow-growth” and “fast-growth” regimes. The particular model pioneered by Hamilton (1989) assumes that output growth fluctuations can be described by an AR(p) process shifting between two states of the economy:10
where S(t) is a discrete random variable that assumes the values 1 or 2 depending on whether the economy is in a slow-growth state (S(t)=1), characterized by a mean growth rate μ1, or a fast-growth state (S(t)=2), characterized by a mean growth rate μ2 (μ2 > μ1). The transition between the two growth states is presumed to be determined by a two-state Markov chain with conditional transition probabilities pij. For example, p11 is the probability that if the economy is in a slow-growth regime in time period t-1, it will again be in a slow-growth regime in time period t. The business cycle would be associated with the alternating shifts in S(t), while the η(t) shocks (propagated through an AR(p) filter) would capture other fluctuations unrelated to the cycle.
14. By contrast to the STPC approach, the RC approach assumes that trend and cycle are fundamentally intertwined phenomena. Moreover, while the STPC approach assumes that expansions and recessions are essentially symmetric mirror images of each other, the RC approach allows for asymmetric durations of expansions and recessions.
D. Empirical Evidence
15. As a first pass at the evidence, Figure I-3 shows the frequency distribution of quarterly real GDP growth data for Switzerland, western Germany, and the United States during the period 1997-98.11 Intuitively, output growth data that are generated by a RC model should cluster around two different growth rates, reflecting the slow- and fast-growth regimes of the economy. The STPC model would, by contrast, predict that output growth should cluster around the average output growth rate of the series. While this approach of examining GDP growth data is unlikely to be conclusive in view of the small sample sizes, the bi-modal appearance of the frequency plot for Swiss GDP growth suggests that at least one of the basic characteristics of regime change models—that the economy switches between two basic growth states—is consistent with the properties of Swiss output data. Similarly, GDP growth data for western Germany appear also to be well described by a bi-modal frequency distribution. Interestingly, there is hardly a trace of bimodality in U.S. GDP growth data, at least for the particular time period chosen for this comparison, although the RC model was originally developed to model business cycle fluctuations in U.S. data.12
Switzerland: Frequency Distribution for Real GDP Growth, 1977.Q2-1998.Q2
(Quarterly growth rates; in percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Switzerland: Frequency Distribution for Real GDP Growth, 1977.Q2-1998.Q2
(Quarterly growth rates; in percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Switzerland: Frequency Distribution for Real GDP Growth, 1977.Q2-1998.Q2
(Quarterly growth rates; in percent)
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
16. More formal statistical evidence is provided by estimation results for the RC model applied to quarterly Swiss GDP data for the period 1977-98:13
This RC model appears to fit Swiss data well. According to the parameter estimates for μ1 and μ2, the economy shifts between a slow-growth regime with average growth of 0.40 percent and a fast-growth state with average growth of 2.3 percent, broadly in line with the frequency distribution evidence shown in Figure I-3. Figure I-4 shows the identified periods of slow growth (recessions) and the unconditional probability of being in a slow-growth state. The estimated probabilities appear to reflect well the recession and boom phases often associated with the cyclical swings of the Swiss economy and also highlight the asymmetric durations of expansions and booms underlying Switzerland’s recent business cycle experience.
Switzerland: A Regime Change Model of Output Fluctuations, 1977.Q2-1998.Q2
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Source: IMF, World Economic Outlook; and staff estimates.Switzerland: A Regime Change Model of Output Fluctuations, 1977.Q2-1998.Q2
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Source: IMF, World Economic Outlook; and staff estimates.Switzerland: A Regime Change Model of Output Fluctuations, 1977.Q2-1998.Q2
Citation: IMF Staff Country Reports 1999, 030; 10.5089/9781451807165.002.A001
Source: IMF, World Economic Outlook; and staff estimates.E. Economic Forces Driving Regime Change Models
17. Descriptive business cycle research in the spirit of Burns and Mitchell (1946) has since long emphasized the division of business cycles into separate regimes or phases. This analysis treated expansions or booms (fast-growth regimes) separately from contractions or recessions (slow-growth regimes). Perhaps reflecting statistical convenience and/or computational constraints, most formal econometric work, however, has until recently focussed on linear one-state models of the business cycle.
18. At this point, the fundamental economic forces that could give rise to cyclical regime changes are not well understood. Recent theoretical research has highlighted that economic activity tends to bunch at certain time frequencies, most conspicuously at daily (day-night), weekly (five-day workweek), and seasonal time frequencies. At these time frequencies, economic activity tends to shift between high- and low-activity regimes. This research has pointed out that the observed bunching of economic activity at these time frequencies is likely to reflect thick-market effects, i.e. economic activity is more efficient when it takes place in a concentrated fashion. Following this lead, Hall (1991) has conjectured that cyclical recessions and booms may in part also reflect persistent periods of slow and high activity related to thick-market effects.14 From this perspective, a recession would represent a period where many producers have an incentive to be relatively inactive as a result of the higher costs of producing related to lower activity of the producer’s customers and suppliers. On the other hand, a boom would occur if some shock energizes many producers at the same time, and the boom becomes self-sustaining because of spillovers and complementarity effects across producers.
19. In the particular case of Switzerland, persistent shifts between slow- and high-growth states at the business cycle frequencies could also reflect factors that are more specific to Switzerland:
First, the supply of labor in Switzerland is relatively elastic, partly reflecting worker migration flows but also large fluctuations in labor force participation rates.15 In this setting, economic booms may have more staying power because output and investment are not as constrained by the available labor force.
Second, the influx of foreign workers may mobilize additional demand for housing, construction projects that would typically have a relatively long gestation period. There may also be strong effects on the demand for durable consumption goods. On the other hand, this process may work in reverse and prolong recessions after the end of a boom period as the influx of workers into the labor force stops and/or foreign workers emigrate.
Third, the supply of loanable funds in Switzerland may be particularly elastic during boom periods, reflecting Switzerland’s high savings rate but also the large and sophisticated financial sector. It is less clear why this factor would help to prolong recessions, although there may be a “hangover effect” on banks’ lending policies after prolonged booms.
Fourth, the pronounced procyclical behavior of fiscal policies in Switzerland may have added to the persistence of booms and recessions.16
F. Implications for Forecasting and Policy Analysis
20. As regards forecasting, regime change models can account for the widely observed persistence of forecast errors after business cycle turning points. As noted in the introduction, a tendency to systematically underpredict (during the 1980s boom) or overpredict (during the 1991-96 recession) has been as conspicuous feature of real GDP growth forecasts for the Swiss economy. Indeed, Hamilton’s (1989) pioneering work on regime change models saw significant improvements in forecasting accuracy as one of the main promises of RC models.
21. It is noteworthy, however, that to the extent that forecasters dislike large forecast revisions, i.e. their loss function depends not only on the size of the forecast error per se but also on the size of the forecast revision per time unit, an RC model may be considered a forecasting tool that implies “excessive” forecast revisions.
22. An RC model would suggest that predicting medium-term output growth (e.g., two-three years ahead) is subject to considerable uncertainties, as the timing of regime shifts may be difficult to predict.17 By contrast, the STPC approach suggests that medium-term output forecasts can be based on the convenient rule-of-thumb that future GDP growth is equal to potential output growth plus an adjustment required to close the output gap over the medium term (usually four-five years). By the same token, to the extent that medium-term inflation forecasts (as constructed, for example, in the context of inflation targeting frameworks) also depend on the projected cyclical state of the economy, these forecasts would likely suffer from similar limitations with regard to forecasting accuracy.
23. Finally, under a regime change model, the size of the automatic response of the fiscal system (automatic fiscal stabilizers) to the state of the economy would depend on the duration of the slow- and fast-growth regime. This would be so because the real growth rates of most public expenditures, apart from unemployment benefits, would likely adjust automatically to a protracted regime shift in the real output growth rate. For example, during a prolonged recession, the real growth rate of public wages would eventually reflect the depressed state of the economy; the same would likely apply for the cost of health care; similarly, the indexation mechanisms of PAYG pension systems would reflect the state of the economy, in particular in the case of (lagged) indexation to wages. The conventional calculation of the size of automatic fiscal stabilizers, however, assumes that structural spending expands at the rate of potential output independently of the duration of booms and recessions.
References
Artis, Michael J., 1997, “How Accurate are the IMF’s Short-Term Forecasts? Another Examination of the World Economic Outlook,” in: Staff Studies for the World Economic Outlook, pp. 1–39 (Washington: International Monetary Fund).
Burns, Arthur F., and Wesley C. Mitchell, 1946, Measuring Business Cycles (New York: National Bureau of Economic Research).
Diamond, Peter, 1982, “Aggregate Demand Management in Search Equilibrium,” Journal of Political Economy, Vol. 90, pp. 881–94.
Diebold, Francis X., and Glenn D. Rudebusch, 1996, “Measuring Business Cycles: A Modern Perspective,” Review of Economics and Statistics, Vol. 78, pp. 67–77.
Hall, Robert E., 1991, Booms and Recessions in a Noisy Economy (New Haven: Yale University Press).
Hamilton, James D., 1989, “A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle,” Econometrica, Vol. 57, No. 2, pp. 357–84.
Harvey, Andrew C., 1989, Forecasting, Structural Time Series Models, and the Kalman Filter (Cambridge: Cambridge University Press).
International Monetary Fund, 1997, Switzerland: Selected Issues and Statistical Appendix, “Estimates of Potential Output Growth and the Cyclical Output Gap,” pp. 7–22, IMF Staff Country Report No. 97/18 (Washington: International Monetary Fund).
International Monetary Fund, 1998, Switzerland: Selected Issues and Statistical Appendix, “Fiscal Policy and the Business Cycle,” pp. 32–57, IMF Staff Country Report No. 98/43 (Washington: International Monetary Fund).
OECD, 1996, Labour Market Policies in Switzerland (Paris: OECD).
Potter, Simon M., 1995, “A Nonlinear Approach to U.S. GNP,” Journal of Applied Econometrics, Vol. 10, pp. 109–25.
Prepared by Albert Jaeger.
The specific version of a regime change model applied to Swiss data in this chapter was pioneered by Hamilton (1989). However, the concept of regime switching at business cycle turning points has a long history as a defining characteristic of business cycles; see e.g. Burns and Mitchell (1946). For a recent overview of empirical business cycle research that highlights the links between the Burns-Mitchell tradition and the modern regime change approach to analyzing business cycles, see Diebold and Rudebusch (1996).
Switzerland’s monetary policy framework is analyzed in Chapter II.
Switzerland’s fiscal policy framework was analyzed in Chapter II of last year’s IMF Staff Country Report No. 98/43.
See the official consultation report on alternative fiscal policy rules under a constitutional balanced budget amendment (Vernehmlassungsbericht zur Schuldenbremse (1995)).
See, e.g., Hall (1991).
Figure I-1 shows one-year ahead real GDP growth forecasts published in the December issues of the OECD Economic Outlook. The forecasting records of other official and private forecasters were similar to that of the OECD.
Chapter I of the IMF Staff Country Report No. 97/18 for Switzerland describes the staff’s approach to measuring potential output using a production function approach.
Provided that all the roots of the autoregressive (AR) polynomial (1 − φL − … − φpLp) the outside the unit circle, the AR(p) process is stationary. In the following, this stationarity assumption will be assumed to hold. However, since the stochastic properties of the time-dependent drift term c(t) are left unrestricted, observed output growth could be nonstationary.
Potter (1995) surveys several different types of regime change models for business cycles.
The time range was restricted to 1977-98 because Swiss GDP growth data underwent a clear structural break in the mid-1970s as regards the level of the mean growth rate. It should also be noted that revised national income accounts data for Swiss real GDP are presently only available since 1980; the GDP growth data before 1980 were spliced with the revised data.
The more recent difficulties of macroeconomic forecasts to track the strength of future U.S. real GDP growth appear, however, to be well accounted for by an RC model.
The routine for estimating the regime change model is written in the GAUSS programming language and was obtained from http://weber.u.washington.edu/~ezivot/econ512.
A formal model of thick-market externalities is developed by Diamond (1982). In his model, the costs of doing business are lower at times of higher aggregate activity.
See OECD (1996).
Chapter II of the IMF Staff Country Report No. 98/43 provides an analysis of the cyclical behavior of fiscal policy at different government levels in Switzerland.
Some recent literature has, however, concluded that business cycles exhibit duration dependence (see Diebold and Rudebusch (1996)). This means in the context of an RC model that, for example, the probability that a slow-growth regime will come to an end increases with the number of time periods spent in the slow-growth regime.
