Predictive Macro-Finance With Dynamic Partition Models

Abstract

Dynamic partition models are used to predict movements in the term structure of interest rates. This allows one to understand historic cycles in the performance of how interest rates behave, and to offer policy makers guidance regarding future expectations on their evolution. Our approach allows for a random number of possible change points in the term structure of interest rates. We use particle learning to learn about the unobserved state variables in a new class of dynamic product partition models that relate macro-variables to term structures. The empirical results, using data from 1970 to 2000, clearly identifies some of the key shocks to the economy, such as recessions. We construct a time series of Bayes factors that, surprisingly, could serve as a leading indicator of economic activity, validated via a Granger causality test. Finally, the in-sample and out-of-sample forecasts from our model are quite robust regardless of the time to maturity of interest rates.

Forecasting the Term Structure of Interest Rates with Potentially Misspecified Models

“…For example, Diebold and Li (2006) find that the three-factor DNS outperforms the RW at 1-month-ahead horizon for short maturities, but for longterm bond yields the RW dominates the DNS. Zantedeschi et al (2011) confirm that the RW forecasts better in the short run whereas at three-and six-step-ahead forecast horizons the predictions from their DNS with time-varying factorloadings are much improved. Moench (2008) finds that the arbitrage free affine term structure model forecasts better than the RW at he six-month maturity only.…”

The predictive density simulation of the yield curve with a zero lower bound

“…Our model can be extended to enhance its predictive accuracy. For instance, one may introduce regime changes to the NDNS(3) model, given the finding of Zantedeschi et al (2011) that a regime-switching DNS model can provide better forecasts for long forecast horizons. In addition, it would be interesting to apply our econometric framework to a global dynamic Nelson-Siegel model to consider cross-country interactions in global bond markets.…”

“…For instance, de Pooter (2007) examines the forecasting ability of DNS models using different numbers of factors and finds that predictive gains can be obtained by adding a second slope factor to the DNS(3) model. Zantedeschi et al (2011) allow for an unknown number of change points in the yield curve dynamics, which helps the model outperform the random-walk model at three-month-ahead and six-month-ahead forecasts.…”

Estimation of state-space models with endogenous Markov regime-switching parameters

“…The 12 months of 2011 are forecasted recursively. I follow Zantedeschi et al (2011) and Chib and Kang (2013), and evaluate the predictive accuracy of the forecasts in terms of the posterior predictive criterion (PPC) of Gelfand and Ghosh (1998). 8 Table 7 presents the result for each model.…”