LOCAL CURRENCY BOND RISK PREMIA IN EMERGING MARKETS: INSIGHTS FROM ADVANCED MACHINE LEARNING TECHNIQUES

Understanding the determinants of local currency bond risk premia is essential for both investors and policymakers in emerging markets, as these premia reflect compensation for risk in economies characterized by heightened volatility and structural heterogeneity.

This study examines the predictive drivers of local currency bond risk premia across six emerging markets—Brazil, Hungary, Poland, Thailand, South Africa, and Turkey—by employing advanced machine learning techniques to capture complex, nonlinear dynamics. The analysis proceeds in two stages: first, it incorporates yield curve-based factors such as forward rates, forward-spot spreads, and term premia; second, it expands the framework to include inflation, implied foreign exchange (FX) volatility, and broader macroeconomic indicators.

The findings reveal pronounced regional divergences: in Turkey, consumer price inflation dominates predictive performance, complemented by macroeconomic variables and FX volatility, while yield curve factors offer little to no explanatory power; in South Africa, implied FX volatility is the leading driver, with inflation and macro indicators playing supporting roles; conversely, in Brazil, Hungary, Poland, and Thailand, yield curve-based predictors consistently outperform inflation, FX volatility, and macroeconomic data.

A range of algorithms—including linear regression, PCA, PLS, neural networks, random forests, XGBoost, and extremely randomized trees—are applied, with ensemble and non-linear methods such as XGBoost, random forests, and neural networks achieving superior predictive accuracy.

By documenting the country-specific nature of predictive determinants and highlighting the utility of combining diverse data sources with modern machine learning, the study advances both empirical asset pricing and risk management in emerging markets, offering practical implications for portfolio allocation, policy design, and the development of more robust frameworks for forecasting excess bond returns.

Project Year: 2025

Project Partner: This is academic research project joint with Emrah Ahi, Levent Güntay and Hasan Taşdemir, based on the MS Thesis of Hasan Taşdemir