Forecasting recessions in real time: Speed Dating with Norwegians

Transcription

1 Forecasting recessions in real time: Speed Dating with Norwegians Knut Are Aastveit 1 Anne Sofie Jore 1 Francesco Ravazzolo 1,2 1 Norges Bank 2 BI Norwegian Business School 12 October 2013

2 Motivation Domenico asked us yestarday: Why do we dating? Example for speeding dating at Norges Bank AJR (Norges Bank) 12 October / 21

3 Interest rate cut in December December 2008: interest rates by 1.75, the biggest cut in more than 22 years. AJR (Norges Bank) 12 October / 21

4 Motivation Point and density forecasts for a few variables might not always be the most relevant information Long tradition in business cycle analysis of separating periods with contraction from periods with expansion Most of the research concentrated on US Extend the business cycle analysis to other countries, in this case Norway, which have not a reference cycle AJR (Norges Bank) 12 October / 21

5 Contribution of this paper What we do Explore alternative methods for defining classical business cycle turning points in Norway A quarterly version of the Bry and Boschan algorithm (Harding and Pagan 2002) An autoregressive Markov switching model (Hamilton 1989) We forecast turning points in real GDP in real time, based on A (nonparametric) extension of the BB based on predictive densities from autoregressive models Parametric extension of the BB based on predictive densities from bivariate VAR based on GDP and financial indices or surveys An autoregressive Markov switching model Autoregressive Markov switching models augmented with financial indices or surveys AJR (Norges Bank) 12 October / 21

6 Contribution of this paper Results Bry Boschan defines a reasonable business cycle in Norwegian economy The Markov switching AR models is, on contrary, not well suited Bry Boschan on forecasts from survey models superior to Markov switching models for predicting turning points in real time AJR (Norges Bank) 12 October / 21

7 Outline 1 Methods for defining classical business cycle turning points 2 Empirical illustrations 3 Predicting turning points Alternative models Empirical exercise Results 4 Summary AJR (Norges Bank) 12 October / 21

8 Classical versus growth cycles AJR (Norges Bank) 12 October / 21

9 Estimating turning points in classical cycles: BB rule The Bry and Boschan rule (BB) (see e.g. Harding and Pagan (2002)) Peak if y t = lngdp t greater than {y t 2, y t 1 } and smaller than {y t+1, y t+2 } ({Pr(y t+1 < y t ) > 0.5} and {Pr(y t+2 < y t ) > 0.5}) Length of expansions and contractions minimum 2 quarters Length of cycle minimum 4 quarters Note: it requires future information to define state at time t AJR (Norges Bank) 12 October / 21

10 Estimating turning points in classical cycles Autoregressive Markov switching (MS) (Hamilton 1989) y t = ν st + φ 1 y t φ p y t p + u t, u t i.i.d. N (0, σ 2 ) t = 1,..., T, where ν st is the MS-intercept; φ l, with l = 1,..., p, are the autoregressive coefficients and {s t } t is the regime-switching process, that is a m-states ergodic and aperiodic Markov-chain process. s t represents the current phase, at time t, of the business cycle. The latent process takes integer values, say s t {1,..., m}. For m = 2, the two states are contraction or expansion. Transition probabilities: P(s t = j s t 1 = i) = p ij, with i, j {1,..., m}. Transition matrix P of the chain: p p 1m P =.. p m1... p mm AJR (Norges Bank) 12 October / 21

11 log GDP Mainland Norway and turning points (BB) Q Q Q Q Q Q Q AJR (Norges Bank) 12 October / 21

12 log GDP Mainland Norway and turning points (Both) Q Q Q Q Q Q AJR (Norges Bank) 12 October / 21

13 Unemployment rate and business cycle dating (BB) based on GDP Q Q Q Q Q Q Q AJR (Norges Bank) 12 October / 21

14 Models for predicting turning points in real time BB requires future information: supplement it with predictive densities from AR for GDP and define recession/expansion when the probability is higher than 50% Markov switching model Relying on just GDP might be not optimal: GDP releases with delay and revised Other variables, in particular financial indicators and survey, might provide useful information BB with forecasts from bivariate indicator models with financial indices and survey data Autoregressive Markov switching models augmented with financial indices and survey data AJR (Norges Bank) 12 October / 21

15 Financial and survey data 1 Financial indices (Naes et al., 2012, and Aastveit and Trovik, 2012): Financial conditions index (FCI) Amihud s illiquidity ratio (Ill) 2 Survey data (Martinsen et al, 2013): Consumer confidence (CC) The overall Confidence Indicator from the Business Tendency Survey (BTS) Norges Bank s Regional Network survey (RN) AJR (Norges Bank) 12 October / 21

16 Empirical exercise We study the latest recession All forecasts (2 steps-ahead) calculated recursively in real time Forecasts are added to the relevant vintage AJR (Norges Bank) 12 October / 21

17 log GDP Mainland Norway. Two vintages GDP Vintage Nov 2008 (left axis) GDP Vintage Feb 2012 (right axis) Q Q Q Q AJR (Norges Bank) 12 October / 21

18 Forecasting turning points in real time - peaks Model date of detection Peak quarter BB with AR(4) May 2008Q3 BB with Consumer confidence (CC) December 2008Q3 BB with Business Tendency Survey (BTS) January 2008Q3 BB with Regional Network Survey (RN) December 2008Q3 BB with Financial conditions index (FCI) December 2008Q3 BB with Amihud s illiquidity ratio (Ill) May 2008Q3 Markov switching with AR(4) May 2008Q2 Markov Switch with CC March 2008Q3 Markov Switch with BTS April 2008Q2 Markov Switch with RN December 2007Q3 Markov Switch with FCI February 2008Q3 Markov Switch with Ill May 2008Q2 AJR (Norges Bank) 12 October / 21

19 Forecasting turning points in real time - troughs Model date of detection Trough quarter BB with AR(4) May 2009Q1 BB with Consumer confidence (CC) June 2009Q1 BB with Business Tendency Survey (BTS) July 2009Q1 BB with Regional Network Survey (RN) June 2009Q1 BB with Financial conditions index (FCI) July 2009Q1 BB with Amihud s illiquidity ratio (Ill) May 2009Q1 Markov switching with AR(4) May 2009Q1 Markov Switch with CC December 2010Q1 Markov Switch with BTS April 2009Q1 Markov Switch with RN December 2008Q2 Markov Switch with FCI February 2008Q4 Markov Switch with Ill August 2009Q1 AJR (Norges Bank) 12 October / 21

20 Forecasting recessions - Main results BryBorschan with AR4 BryBorschan with Consumer Confidence BryBorschan with Business tendency survey BryBorschan with Regional network survey BryBorschan with Financial conditions index (FCI) BryBorschan with Amihud's illiquidity index (Ill) Markov Switcing with AR4 Markov Switcing with with Consumer Confidence Markov Switcing with Business tendency survey Markov Switcing with Regional network survey Markov Switcing with Financial conditions index (FCI) Markov Switcing with Amihud's illiquidity index (Ill) Markov Switcing with unemployment (registered) AJR (Norges Bank) 12 October / 21

21 Summary Try to define and forecast classical business cycles in Norway using alternative methods and models Business cycles defined very differently depending on the method used Bry Boschan rule seems to work well Markov switching works less well Difficult to pinpoint turning points in real time GDP are revised, often substantially Forecasts from financial and survey data helps AJR (Norges Bank) 12 October / 21