Abstract
The macro-financial data are characterized by heteroskedasticity which leads to inconsistent estimates and inference from binary choice models (BCMs). To address this problem, we propose a generalized autoregressive conditionally heteroskedastic-type adjustment for the conditional variance of model errors. A data augmentation type algorithm is developed for estimation while Lagrange multiplier (LM)-type tests are derived for testing ARCH effects in BCMs. Simulation results show that the proposed model leads to bias reduction in estimates, while the expected Information matrix-based LM test exhibits smaller size distortions and higher power properties. Empirically, predictions of the US business and financial cycles reaffirm the effectiveness of the extended model.
Original language | English |
---|---|
Pages (from-to) | 2007-2035 |
Number of pages | 29 |
Journal | Journal of Statistical Computation and Simulation |
Volume | 86 |
Issue number | 10 |
DOIs | |
Publication status | Published - 2 Jul 2016 |
Keywords
- Binary choice models
- heteroskedastic-probit
- heteroskedasticity
- ARCH LM test
- probit-GARCH
- 91B84
- 62M10
- 91B55
- PREDICTING US RECESSIONS
- STOCK-MARKET
- LEADING INDICATORS
- RESPONSE MODELS
- TERM STRUCTURE
- PROBIT MODELS
- HETEROSCEDASTICITY
- BANK
- VARIABLES
- VOLATILITY