Leading Indicators in a Globalised World

1
Leading Indicators in a Globalised World

• F. Fichtner, R. Rüffer, B. Schnatz
• Rome, 27 March 2009

2
Introduction

• Motivation
• Leading indicators Traditional and important
  tool in applied business cycle analysis.
• Potential shortcoming Mainly based on domestic
  variables.

3
Introduction

• Research questions
• How useful are leading indicators for projecting
  economic activity?
• Has the performance of leading indicator models
  diminished over time owing to globalisation?
• Does the inclusion of foreign leading indicators
  improve forecast accuracy?

3
4
Data

• OECD Composite Leading Indicator (CLI)
• Component series are selected on a
  country-by-country basis
• Economic reason for leading relationship
• Leading indicator cycle leads that of the
  reference series
• Reasonable data quality (timeliness, revisions)
• Reference Series Industrial Production
• Sample 1975m1 - 2008m4
• 11 countries (plus OECD series)
• United States, United Kingdom, Japan, Canada,
  Denmark, Sweden.
• Germany, Spain, France, Greece and Italy.

5
Stylised Facts
CLI and Industrial production in OECD countries
Cross-correlation of CLI and industrial production
6
Empirical strategy and dimensions

• Linear versus non-linear models
• In-sample versus out-of-sample forecast
  evaluation ?
• Iterated multi-step forecasts versus direct
  forecasts ?
• Iterated MS forecasts
• Direct forecasts
• Accounting for integrated variables VAR versus
  VEC-models ?

6
7
Forecast evaluation

• Diebold-Mariano (1995) test with small sample
  correction suggested by Harvey et al. (1997)
• Benchmark model Univariate autoregression, 15
  year rolling window.
• Lag selection incl. 8 lags j lt 24 with the
  highest significance.

7
8
Forecast performance of domestic CLI

• VAR models provide better forecasts than
  benchmark.
• Sizeable fraction of the results is statistically
  significant.
• Major gains in medium-term horizon and for larger
  countries.

9
Value added of cointegration

• Industrial production and CLI are I(1) variables.
• Accounting for cointegration further improves
  forecast accuracy (mostly significantly).
• Improvements are less pronounces over longer
  horizons puzzle?

10
Iterative versus direct forecasts

• Ambiguous results.
• Direct forecasts do not improve the accuracy of
  forecasts over iterative forecasts.

10
11
Have forecasting properties diminished?

• Compute for each period an average forecast (h
  1,12).? Information on the quality of all
  forecasts computed in one particular period.
• As forecast errors are larger for longer forecast
  horizons ? Normalise errors with the respective
  standard deviation.
• Output volatility changes over time (great
  moderation) ? Analyse difference between (2)
  and forecasts over h- horizons based on a
  univariate AR-model.? Negative values VEC
  outperforms UAR.
• Evolution over time Illustrate trends with
  regression line.

11
12
Performance over time (VEC vs. UAR)
13
International Dimension

• Benchmark VEC,
• For CLIEXT, we consider five proxies
• CLI for the OECD aggregate (OTO)
• CLI for the United States (USA)
• Trade-weighted CLI (EXT)
• Replace the domestic CLI (save degrees of
  freedom)
• with average of the domestic CLI and all foreign
  CLIs (simple averaging),
• with first principal component of the domestic
  CLI and all foreign CLIs.

14
Relative Performance of International CLI
15
Forecast combination
15
16
Conclusion

• CLI includes useful information for projecting IP
• CLI model significantly outperforms (naïve)
  univariate model.
• accounting for co-integration improves forecast
  substantially,
• direct forecasts do not improve over iterative
  forecasts.
• But relative forecast performance of CLI model
  seems to decrease over time ? globalisation?
• Inclusion of external CLIs provides mixed
  results.
• Forecast combination is promising avenue.

17

• THE END

17
18
International CLI Over Time