Optimizing Policymakers’ Loss Functions in Crisis Prediction: Before, Within or After?
Early-warning models most commonly optimize signaling thresholds on crisis probabilities. The expost threshold optimization is based upon a loss function accounting for preferences between forecast errors, but comes with two crucial drawbacks: unstable thresholds in recursive estimations and an in-sample overfit at the expense of out-of-sample performance. We propose two alternatives for threshold setting: (i) including preferences in the estimation itself and (ii) setting thresholds ex-ante according to preferences only. Given probabilistic model output, it is intuitive that a decision rule is independent of the data or model specification, as thresholds on probabilities represent a willingness to issue a false alarm vis-à-vis missing a crisis. We provide simulated and real-world evidence that this simplification results in stable thresholds and improves out-of-sample performance. Our solution is not restricted to binary-choice models, but directly transferable to the signaling approach and all probabilistic early-warning models.
IWH FDI Micro Database
IWH FDI Micro Database The IWH FDI Micro Database (FDI = Foreign Direct...
The Joint Dynamics of Sovereign Ratings and Government Bond Yields
Journal of Banking & Finance,
Can a negative shock to sovereign ratings invoke a vicious cycle of increasing government bond yields and further downgrades, ultimately pushing a country toward default? The narratives of public and political discussions, as well as of some widely cited papers, suggest this possibility. In this paper, we will investigate the possible existence of such a vicious cycle. We find no evidence of a bad long-run equilibrium and cannot confirm a feedback loop leading into default as a transitory state for all but the very worst ratings. We use a bivariate semiparametric dynamic panel model to reproduce the joint dynamics of sovereign ratings and government bond yields. The individual equations resemble Pesaran-type cointegration models, which allow for valid interference regardless of whether the employed variables display unit-root behavior. To incorporate most of the empirical features previously documented (separately) in the literature, we allow for different long-run relationships in both equations, nonlinearities in the level effects of ratings, and asymmetric effects in changes of ratings and yields. Our finding of a single good equilibrium implies the slow convergence of ratings and yields toward this equilibrium. However, the persistence of ratings is sufficiently high that a rating shock can have substantial costs if it occurs at a highly speculative rating or lower. Rating shocks that drive the rating below this threshold can increase the interest rate sharply, and for a long time. Yet, simulation studies based on our estimations show that it is highly improbable that rating agencies can be made responsible for the most dramatic spikes in interest rates.
Qual VAR Revisited: Good Forecast, Bad Story
Journal of Applied Economics,
Due to the recent financial crisis, the interest in econometric models that allow to incorporate binary variables (such as the occurrence of a crisis) experienced a huge surge. This paper evaluates the performance of the Qual VAR, originally proposed by Dueker (2005). The Qual VAR is a VAR model including a latent variable that governs the behavior of an observable binary variable. While we find that the Qual VAR performs reasonable well in forecasting (outperforming a probit benchmark), there are substantial identification problems even in a simple VAR specification. Typically, identification in economic applications is far more difficult than in our simple benchmark. Therefore, when the economic interpretation of the dynamic behavior of the latent variable and the chain of causality matter, use of the Qual VAR is inadvisable.
The Performance of Short-term Forecasts of the German Economy before and during the 2008/2009 Recession
International Journal of Forecasting,
The paper analyzes the forecasting performance of leading indicators for industrial production in Germany. We focus on single and pooled leading indicator models both before and during the financial crisis. Pairwise and joint significant tests are used to evaluate single indicator models as well as forecast combination methods. In addition, we investigate the stability of forecasting models during the most recent financial crisis.