Leading indicators of the business cycle: dynamic probit models for OECD countries and Russia

Transcription

1 18th International Conference on Macroeconomic Analysis and International Finance Leading indicators of the business cycle: dynamic probit models for OECD countries and Russia Anna Pestova Research fellow, NRU-HSE, IEF RAS Senior expert, CMASF 1 Rethymno, Greece May 31, 2014

2 Outline 1. Motivation and objective 2. Literature review 3. Methodology and data 4. Estimation results 5. Conclusion 2 2

3 Motivation and objective End 1990s mid 2000s: «end of the business cycle» - Weber (1997), «great moderation» - Stock, Watson (2002) growth of service sector, technological shifts, globalization of production, change in monetary policy End of 2000s «great recession»: all major European countries were in recession Today risks of renewed recession remain high (for many OECD countries risks of its continuation) The aim of the paper is to assess whether it is possible to predict the state of the business cycle (particularly, recessions) in advance Leading indicators of the business cycle phase change Panel dataset of OECD countries and Russia 3 Different forecasting horizons (up to 1 year)

4 Literature review 2 directions of research on leading indicators Continuous dependent variable DFM models - Stock, Watson (1989, 2006), Forni et al. (2001, 2003), etc. Nonmodel based LI - OECD (2008) Discrete dependent variable Binary models - Estrella, Mishkin (1998), Stock, Watson (1992), Kauppi, Saikkonen (2008), Ng (2012), etc. Contribution Dynamic panel data models (previous studies time series data) Credit market variables as leading indicators 9 Longer forecasting horizon (up to 1 year vs several months in existing studies)

5 Methodology and data 9 Panel quarterly data on ~30 OECD countries and peers (including Russia) over the period 1q1980 2q2013*. Data sources IFS, OECD, WB, ECB Dependent variable state of the economy / business cycle phase (binary) Methodology: dynamic panel probit model * y it y it k in Dueker (1997), Moneta (2003), Kauppi, Saikkonen (2008), Ng (2012) dynamic probit model on time series data * y it y it xit k y it k x it k u * 1if y 0 it 0 else latent dependent variable observed binary variable vector of explanatory variables i number of quarter lag (1, 2 and 4 quarter) it According to the data, initial condition is exogenous with respect to individual effects : no «initial condition problem» - standard RE probit estimator is used i * GDP contraction during transitional period in post socialist countries was deleted from the sample y i1

6 Dating business cycle phases 3 approaches Classical business cycle (peaks and troughs) Deviation cycle Growth rate cycle Target variable GDP growth rates over corresponding quarter of the previous year (in the literature - IIP, CCI, etc.) 1 if GDPgr 0 y it 0 f GDPgr 0 (recession ) (expansion ) 10

7 Explanatory variables selection 1. Macroeconomic variables lagged GDP growth, inflation, unemployment, consumption and investment expenditures, inventories, housing market dynamics 2. Consumer and business expectations 3. External sector variables US business cycle phase, oil prices. 4. Financial sector variables government bond and interbank market interest rates, stock market indices, monetary aggregates, NEER 5. Credit market variables proposed in the paper lending rates credit boom indicators 7 (1) (4) on the basis of Estrella, Mishkin (1998), Stock, Watson (1993), Birchenhall et al., (1999), Ozildirim et al. (2010), Castro (2010), Kauppi, Saikkonen (2008), Ng (2012)

8 8 Estimation results Lag = 1 quarter Lag = 2 quarters Lag = 4 quarters Dependent variabe (lag = k quarters) 1.726*** 0.910*** 0.170* Macroeconomic variables Investment in fixed capital, growth rates per year *** *** CPI inflation, per year 0.034*** Expectations Consumer confidence indicator, in OECD methodology, growth rate per quarter *** *** *** External sector variables US GDP leading indicator, in OECD methodology, in annual terms *** *** *** Current account balance to GDP ratio * *** REER index, 2005= ** 0.009*** Financial sector variables Stock price index, growth rates per quarter *** *** *** Credit market variables Loans to deposits ratio 0.002*** 0.002* 0.006*** Constant *** *** *** Number of observations (countries ) 2260 (25) 2250 (28) 2171(28) LR-test, absence of random effects (P-value) Recessions, correctly classified 86% 81% 70% Absence of recessions, correctly classified 90% 85% 79% Noise-to-signal ratio 12% 19% 31% Примечание: ***, ** и * - значимость коэффициентов на 1, 5 и 10 уровне.

9 Discussion of the results Lags of the dependent variable do matter (dynamic mechanism), however, their influence weakens with lag increase Credit market variables help to predict phase of the BC. High LTD ratio means proximity to the peak of the credit and, with a lag, business cycle Slowdown of investment, deterioration of consumer confidence and inflationary pressure are the main nonfinancial domestic indicators External sector indicators are significant: phase of US BC, CA balance and excessive currency appreciation Stock market contraction leads turning points of the BC 9 Quality of in-sample prediction decreases with lag increase However, even 4 quarter leading indicators model could be considered as satisfactory (70% recession were predicted correctly with N/S about 30%)

10 Conclusion In this paper leading indicators of the business cycle phase were first in known empirical literature constructed on the panel data set Quarterly data, growth rates cycle Dynamic nature of the business cycle was captured Different forecasting horizons (up to 1 year) were tested. Best predictions were achieved on 1 quarter lag model (about 90% of observations were correctly classified). Prediction results with year lag were less precise (more than 70% of correct classification with 30% noise ratio) but still useful 10 Credit market variables proposed in the paper performed well as leading indicators in all tested models

11 Further directions of work Check robustness of the results Calculate out-of-sample predictive power Compare current probabilities of recession in sample countries with macroeconomic forecast of IMF (WEO) and leading indicators of OECD 11

12 Thank you for your attention! Anna Pestova 12

13 13 Appendix. Pairwise models Lag = 1 quarter Lag = 2 quarters Lag = 4 quarter Dependent variabe (lag = K quarters) 2.054*** 1.490*** 0.525*** Macroeconomic variables GDP growth rates, per year *** *** Consuption expenditures, growth rates per year *** *** *** Investment in fixed capital, growth rates per year *** *** *** Inventories to GDP ratio, growth per year *** *** ** Unemployment rate ** *** CPI inflation, per year * Expectations Consumer confidence indicator in OECD methodology, growth rate per quarter *** *** *** Business confidence indicator in OECD methodology, growth rate per quarter *** *** *** External sector variables US GDP leading indicator, in OECD methodology, in annual terms *** *** *** Oil price, growth rate per quarter * * Current account balance to GDP ratio *** *** REER index, 2005= *** 0.009*** 0.011*** Financial sector variables Stock price index, growth rates per quarter *** *** *** Government bond interest rate 0.012* 0.012* Credit market variables Spread between lending rate and government bond interest rate 0.021* 0.031*** 0.039*** Spread between money market rate and government bond interest rate *** 0.032** Loans to deposits ratio 0.002*** 0.002*** 0.003*** Loans to GDP ratio 0.002*** 0.002*** 0.003***