Business cycle dating and forecasting with real-time Swiss GDP data


We develop a small-scale dynamic factor model for the Swiss economy allowing for nonlinearities by means of a two-state Markov chain. The selection of an appropriate set of indicators utilizes a combinatorial algorithm. The model’s forecasting performance is as good as that of peers with richer dynamics. It proves particularly useful for a timely assessment of the business cycle stance, as the recessionary regime probabilities tend to have a leading property. The model successfully anticipated the downturn of the 2008–2009 recession and promptly indicated a fall in GDP growth following the discontinuation of the exchange rate floor of the Swiss Franc.

This is a preview of subscription content, log in to check access.

Fig. 1
Fig. 2
Fig. 3
Fig. 4


  1. 1.

    To the surprise of markets and institutions, the Swiss National Bank (SNB) decided for a discontinuation of the Euro-Swiss Franc (CHF) on January 15, 2015, which had been introduced on September 6, 2011.

  2. 2.

    Applications of such linear models to countries include for instance Argentina, Canada, Czech Republic, Spain, Switzerland, etc. (Chernis and Sekkel 2017; Camacho and Pérez-Quirós 2011; Camacho et al. 2015a; Rusnák 2016; Marcellino et al. 2016; Galli 2018).

  3. 3.

    See for instance Giannone et al. (2008), Rünstler et al. (2009), Barhoumi et al. (2010), Marcellino and Schumacher (2010), Bańbura and Rünstler (2011), Aastveit and Trovik (2012), and D’Agostino et al. (2012); see Bańbura et al. (2010) for a review.

  4. 4.

    Camacho et al. (2015b) show that this one-step estimation outperforms the estimation of a Markov-switching process on the factor in a second step.

  5. 5.

    See for instance Faust and Wright (2013), Umer et al. (2018), among others.

  6. 6.

    See for instance Stock and Watson (1999).

  7. 7.

    Alternatively, one could also consider three states as for instance in Carstensen et al. (2017), who distinguish between normal and severe recessions in an application to the German economy.

  8. 8.

    Since the evolution of macroeconomic series is smooth enough, such an approximation is appropriate. For instance, Proietti and Moauro (2006) avoid this approximation at the cost of moving to a complicated nonlinear state-space model.

  9. 9.

    In Switzerland, two distinct authorities are responsible for quarterly and yearly GDP estimates. Based on the yearly GDP measures from the Federal Statics Office (FSO), the State Secretariat of Economic Affairs (SECO) uses temporal disaggregation methods to estimate quarterly GDP figures, which are published periodically about 65 days after the end of a quarter. Revisions to the real-time measure from SECO can stem from different sources: (1) revisions to quarterly indicators; (2) revisions to annual base data; (3) changes in the methodology of national accounts (benchmark revisions); (4) minor changes in the quarterly estimation methods; (5) technical reasons like changes in seasonal adjustment.

  10. 10.

    We start the revision analysis for the vintage 2004-Q1 up to the final vintage 2016-Q4. Our results are qualitatively robust for using different vintages of \(y_t^{f}\).

  11. 11.

    The terminology as regards soft versus hard indicators is not necessarily restricted to truly soft or hard indicators. In fact, if a hard indicator, as for instance industrial production, were to be used in the model as year-over-year growth rate, then this transformed variable would have to be treated in the model as a soft indicator.

  12. 12.

    The AR(2) assumption can be considered as a parsimonious specification: (1) It only requires the estimation of two parameters; (2) it allows a rich dynamic pattern since the roots of \(\phi _q(L)\) and \(\varvec{\Phi }_u(L)\) can be complex.

  13. 13.

    We omitted autoregressive terms in Eq. (5) as they turned out to be not significantly different from zero.

  14. 14.

    Means, medians or zeros are valid alternatives.

  15. 15.

    Technical details are explained in online Appendix.

  16. 16.

    In the case of Stock and Watson (1991), they chose the four monthly coincident variables comprised in the Index of Coincident Economic Indicators (CEI) compiled by the US Department of Commerce (DOC). In particular industrial production, total personal income less transfer payments, total manufacturing and trade sales and employees on non-agricultural payrolls.

  17. 17.

    Imports, exports, overnight stays, retail sales, new car registrations, energy consumption, term spread, Swiss market index (SMI), Swiss performance index (SPI), oil price, real and nominal effective exchange rate, bank assets, loans, KOF industrial orders, PMI, UBS consumer survey, KOF industry and construction surveys, vacancy postings, unemployment rate, social security contributions, CPI, EPI, IFO survey, ZEW survey.

  18. 18.

    In principle, we could also choose \(k=3\) or \(k=5\). We chose \(k=4\) keeping in mind that the DSFM of Stock and Watson (1991) was built on the same number of indicators.

  19. 19.

    With our approach, the maximum number of variables being considered in the selection algorithm is fixed—this can indeed weigh on the fit of the model. In Sect. 5.2 we show in how far the inclusion of further variables changes the fit of our final-model.

  20. 20.

    Optimally, we would choose the same variables as in Stock and Watson (1991). For Switzerland, however, such data do not exist, either because of the frequency (e.g., employment is only available on a quarterly frequency) or lack of data (e.g., industrial production).

  21. 21.

    Both indicators exhibit higher correlation with the year-on-year GDP growth rate than with quarter-on-quarter rate, which is why they load with 11 lags on the common factor. The model outcome does not change qualitatively when they are specified as hard indicators, i.e., loading contemporaneously on the factor.

  22. 22.

    An interesting alternative to stock market volatility would be a general measure of financial market stress (Duprey et al. 2017; Glocker and Kaniovski 2014) or a measure of business uncertainty (Glocker and Hölzl 2019); unfortunately data for these measures are not available.

  23. 23.

    The information used in the model stems entirely from business cycle indicators. Economic policy as such does not enter. However, our model is flexible enough so that it could be extended to combine the following two pieces of information on economic policy in real time: (i) the ex-ante path of policy as published/announced by policy makers; (ii) incoming, observed data on the actual degree of implementation of ongoing plans. In this context Pérez-Quirós et al. (2015), Riguzzi and Wegmueller (2015), Glocker (2013, 2012), among others, show that government (consumption) spending conveys useful information about ex-post policy developments relevant for GDP.

  24. 24.

    The calculations are explained in detail in online Appendix.

  25. 25.

    We consider our decision rule as rather agnostic—a commonly used alternative threshold is a value of 0.5 [see for instance (Hamilton 1989; Carstensen et al. 2017)]. According to our results, the choice of threshold is of second-order importance, as the smoothed state probabilities quickly jumps to one whenever a technical recession materialized.

  26. 26.

    ECRI classifies an episode as a recession in which companies dismiss employees, incomes fall, spending goes down, and output declines—the co-movement of all four variables is key. According to Lakshman and Banerji (2004), this definition provides clarity when it comes to determining if a recession has begun, unlike the popular “two quarters of negative GDP growth” rule of thumb, according to which, if GDP falls for two straight quarters, we have met the “technical” definition of a recession. GDP is just a measure of an economy’s output. But if employment, income, and sales do not fall at the same time, the temporary period of negative-output growth will not catch on and spread, and no recession will occur.

  27. 27.

    For completeness, we have also compared our recession estimates to the business cycle dates published by the OECD. Rather than recessions, this approach identifies the time between a business cycle peak and trough. The OECD business cycle phases are based on the growth-cycle approach, where cycles and turning points are measured and identified in the deviation from trend-series. Against this background, the OECD business cycle phases comprise only a vague basis of comparison. Nevertheless, our recession probabilities are well in line with the identified downswings—we would like to thank an anonymous referee who pointed this out.

  28. 28.

    An evaluation of forecast errors by using the ex-post data for a specific point in time is questionable since measures of forecast errors—as root-mean-squared error (RMSE)—can be deceptively lower when using ex-post data for GDP rather than real-time data (Stark and Croushore 2002).

  29. 29.

    To the best of our knowledge, there is no real-time data of monthly Swiss economic indicators publicly available. Of the ten monthly indicators, only imports and sales might have undergone substantial revisions. Financial variables are not revised, and revisions to survey data are seldom and at most marginal.

  30. 30.

    Diebold and Mariano (1995) provide a pairwise test to analyze whether the differences between two or more competing models are statistically significant. As there is potentially a short-sample problem, we apply the modified version of the Diebold–Mariano test according to Harvey et al. (1997).

  31. 31.

    Participants of the meeting are the State Secretariat for Economic Affairs (SECO), the Federal Customs Administration (FCA), the Swiss Federal Statistical Office (FSO), the Federal Finance Administration (FFA) and the Swiss National Bank (SNB).

  32. 32.

    For instance major banks or economic research institutes, among others.

  33. 33.

    Consider online Appendix for the technical details on the calculation of annual GDP growth rates from quarterly growth rates.

  34. 34.

    We have omitted the confidence bands for better visibility of the point estimates.

  35. 35.

    In this context, news does not refer to data revisions as in Sect. 2.2, but rather to economic sentiment and the surprises therein.

  36. 36.

    The results for this are available upon request.

  37. 37.

    This implies that Eq. (16) comprises two independent autoregressive processes. We maintain a lag-order of two for all lag polynomials. The state-space representation of the two-factor model comprises a straight forward extension to the one of the single-factor model outlined in online Appendix. Details on this and on the estimated coefficients of the model are available upon request.


  1. Aastveit K, Trovik T (2012) Nowcasting norwegian GDP: the role of asset prices in a small open economy. Empir Econ 42(1):95–119

    Google Scholar 

  2. Adrian T, Shin HS (2010) Liquidity and leverage. J Financial Intermed 19(3):418–437

    Google Scholar 

  3. Ang A, Piazzesi M, Wei M (2006) What does the yield curve tell us about GDP growth? J Econom 131(1–2):359–403

    Google Scholar 

  4. Aruoba SB (2008) Data revisions are not well behaved. J Money Credit Bank 40(2–3):319–340

    Google Scholar 

  5. Bai J, Ng S (2008) Forecasting economic time series using targeted predictors. J Econom 146(2):304–317

    Google Scholar 

  6. Bańbura M, Rünstler G (2011) A look into the factor model black box: publication lags and the role of hard and soft data in forecasting GDP. Int J Forecast 27(2):333–346

    Google Scholar 

  7. Bańbura M, Giannone D, Reichlin L (2010) Large Bayesian vector auto regressions. J Appl Econom 25(1):71–92

    Google Scholar 

  8. Barhoumi K, Darné O, Ferrara L (2010) Are disaggregate data useful for factor analysis in forecasting French GDP? J Forecast 29(1–2):132–144

    Google Scholar 

  9. Boivin J, Ng S (2006) Are more data always better for factor analysis? J Econom 132(1):169–194

    Google Scholar 

  10. Calhoun G, Elliott G (2012) Why do nonlinear models provide poor macroeconomic forecasts? In: Seventh ECB workshop on forecasting techniques-new directions for forecasting, Frankfurt am Main, Germany

  11. Camacho M, Pérez-Quirós G (2010) Introducing the euro-sting: short-term indicator of euro area growth. J Appl Econom 25(4):663–694

    Google Scholar 

  12. Camacho M, Pérez-Quirós G (2011) Spain-sting: Spain short-term indicator of growth. Manch Sch 79(s1):594–616

    Google Scholar 

  13. Camacho M, Garcia-Serrador A (2014) The euro-sting revisited: the usefulness of financial indicators to obtain euro area GDP forecasts. J Forecast 33(3):186–197

    Google Scholar 

  14. Camacho M, Dal Bianco M, Martinez-Martin J (2015a) Toward a more reliable picture of the economic activity: an application to Argentina. Econ Lett 132:129–132

    Google Scholar 

  15. Camacho M, Pérez-Quirós G, Poncela P (2015b) Extracting nonlinear signals from several economic indicators. J Appl Econom 30(7):1073–1089

    Google Scholar 

  16. Camacho M, Pérez-Quirós G, Poncela P (2018) Markov-switching dynamic factor models in real time. Int J Forecast 34(4):598–611

    Google Scholar 

  17. Carstensen K, Heinrich M, Reif M, Wolters MH (2017) Predicting ordinary and severe recessions with a three-state markov-switching dynamic factor model. An application to the German business cycle. CESifo working paper series 6457, CESifo Group Munich

  18. Chauvet M (1998) An econometric characterization of business cycle dynamics with factor structure and regime switching. Int Econ Rev 39(4):969–996

    Google Scholar 

  19. Chernis T, Sekkel R (2017) A dynamic factor model for nowcasting Canadian GDP growth. Empir Econ 53(1):217–234

    Google Scholar 

  20. D’Agostino A, McQuinn K, O’Brien D (2012) Nowcasting Irish GDP. OECD J: J Bus Cycle Meas Anal 2012(2):21–31

    Google Scholar 

  21. DeJong DN, Liesenfeld R, Richard J-F (2005) A non-linear forecasting model of GDP growth. Rev Econ Stat 87(4):697–708

    Google Scholar 

  22. Diebold FX, Mariano RS (1995) Comparing predictive accuracy. J Bus Econ Stat 13(3):253–263

    Google Scholar 

  23. Diebold FX, Rudebusch GD (1996) Measuring business cycles: a modern perspective. Rev Econ Stat 78(1):67–77

    Google Scholar 

  24. Duprey T, Klaus B, Peltonen T (2017) Dating systemic financial stress episodes in the EU countries. J Financial Stab 32(C):30–56

    Google Scholar 

  25. Faust J, Wright JH (2013) Forecasting inflation. In: Elliott G, Timmermann A (eds) Part A of handbook of economic forecasting, vol 2. Elsevier, Amsterdam, pp 2–56

    Google Scholar 

  26. Galli A (2018) Which indicators matter? Analyzing the Swiss business cycle using a large-scale mixed-frequency dynamic factor model. J Bus Cycle Res 14(2):179–218

    Google Scholar 

  27. Galli A, Hepenstrick C, Scheufele R (2017) Mixed-frequency models for tracking short-term economic developments in Switzerland. Working papers 2017-02, Swiss National Bank

  28. Giannone D, Reichlin L, Small D (2008) Nowcasting: the real-time informational content of macroeconomic data. J Monet Econ 55(4):665–676

    Google Scholar 

  29. Glocker C (2012) Unemployment compensation and aggregate fluctuations. Int Rev Econ 59(1):21–39

    Google Scholar 

  30. Glocker C (2013) Government expenditures and business cycles—policy reaction and surprise shocks. J Appl Econ Res 7(3):215–254

    Google Scholar 

  31. Glocker C, Hölzl W (2019) Assessing the economic content of direct and indirect business uncertainty measures. WIFO working papers 576, Austrian Institute of Economic Research

  32. Glocker C, Kaniovski S (2014) A financial market stress indicator for Austria. Empirica 41(3):481–504

    Google Scholar 

  33. Hamilton JD (1989) A new approach to the economic analysis of non-stationary time series and the business cycle. Econometrica 57(2):357–384

    Google Scholar 

  34. Harvey D, Leybourne S, Newbold P (1997) Testing the equality of prediction mean squared errors. Int J Forecast 13(2):281–291

    Google Scholar 

  35. Indergand R, Leist S (2014) A real-time data set for Switzerland. Swiss J Econ Stat 150(4):331–352

    Google Scholar 

  36. Kim C-J (1994) Dynamic linear models with Markov-switching. J Econom 60(1–2):1–22

    Google Scholar 

  37. Kim C-J, Yoo J-S (1995) New index of coincident indicators: a multivariate Markov switching factor model approach. J Monet Econ 36(3):607–630

    Google Scholar 

  38. Lakshman A, Banerji A (2004) Beating the business cycle: how to predict and profit from turning points in the economy. Crown Business, Midtown Manhattan

    Google Scholar 

  39. Mankiw NG, Shapiro MD (1986) News or noise? An analysis of GNP revisions. Technical report 1939, National Bureau of Economic Research, Inc

  40. Marcellino M, Schumacher C (2010) Factor MIDAS for nowcasting and forecasting with ragged-edge data: a model comparison for German GDP*. Oxf Bull Econ Stat 72(4):518–550

    Google Scholar 

  41. Marcellino M, Porqueddu M, Venditti F (2016) Short-term GDP forecasting with a mixed-frequency dynamic factor model with stochastic volatility. J Bus Econ Stat 34(1):118–127

    Google Scholar 

  42. Mariano RS, Murasawa Y (2003) A new coincident index of business cycles based on monthly and quarterly series. J Appl Econom 18(4):427–443

    Google Scholar 

  43. Nierhaus W, Abberger K (2015) ifo konjunkturampel revisited. Ifo Schnelldienst 68(05):27–32

    Google Scholar 

  44. Pérez-Quirós G, Pérez JJ, Paredes J (2015) Fiscal targets. A guide to forecasters? Working paper series 1834, European Central Bank

  45. Proietti T, Moauro F (2006) Dynamic factor analysis with non-linear temporal aggregation constraints. J R Stat Soc: Ser C (Appl Stat) 55(2):281–300

    Google Scholar 

  46. Riguzzi M, Wegmueller P (2015) Economic openness and fiscal multipliers. Diskussionsschriften dp1504, Universitaet Bern, Departement Volkswirtschaft

  47. Rünstler G, Barhoumi K, Benk S, Cristadoro R, Den Reijer A, Jakaitiene A, Jelonek P, Rua A, Ruth K, Van Nieuwenhuyze C (2009) Short-term forecasting of GDP using large datasets: a pseudo real-time forecast evaluation exercise. J Forecast 28(7):595–611

    Google Scholar 

  48. Rusnák M (2016) Nowcasting Czech GDP in real time. Econ Model 54(C):26–39

    Google Scholar 

  49. Stark T, Croushore D (2002) Forecasting with a real-time data set for macroeconomists. J Macroecon 24(4):507–531

    Google Scholar 

  50. Stock JH, Watson MW (1991) A probability model of the coincident economic indicators. Cambridge University Press, Cambridge, pp 63–90

    Google Scholar 

  51. Stock JH, Watson MW (1999) A comparison of linear and nonlinear univariate models for forecasting macroeconomic time series. Oxford University Press, Oxford, pp 1–44

    Google Scholar 

  52. Umer UM, Sevil T, Sevil G (2018) Forecasting performance of smooth transition autoregressive (star) model on travel and leisure stock index. J Finance Data Sci 4(2):90–100

    Google Scholar 

  53. Wheelock DC, Wohar ME (2009) Can the term spread predict output growth and recessions? A survey of the literature. Fed Reserve Bank St. Louis Rev 91(5 Part 1):419–440

    Google Scholar 

Download references

Author information



Corresponding author

Correspondence to Christian Glocker.

Ethics declarations

Conflict of Interest

The authors declare that they have no conflict of interest.

Human participants or animals

This article does not contain any studies with human participants or animals performed by any of the authors.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

The authors would like to thank the editor (Robert M. Kunst), two anonymous referees, Ferdy Adam, Lionel Fontagne, Alain Galli, Massimiliano Marcellino and Gabriel Pérez-Quirós for valuable comments. P. Wegmueller would like to thank Ronald Indergand for support. Thanks also goes to seminar participants at SECO/Berne, WIFO/Vienna, SNB/Zurich, STATEC/Luxembourg, the SSES and ifo Dresden conferences for valuable comments. We are very grateful to Astrid Czaloun for research assistance. All errors are our own responsibility. The views expressed in this paper are those of the authors and do not necessarily represent those of the Swiss State Secretariat for Economic Affairs (SECO).

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary material 1 (pdf 359 KB)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Glocker, C., Wegmueller, P. Business cycle dating and forecasting with real-time Swiss GDP data. Empir Econ 58, 73–105 (2020).

Download citation


  • Dynamic factor model
  • Nowcasting
  • Real-time data
  • Markov-switching
  • Business cycle dating

JEL classification

  • C32
  • C53
  • E37