[Paper Review] Consistent Long-Term Yield Curve Prediction

Josef Teichmann, Mario V. Wüthrich|arXiv (Cornell University)|Mar 9, 2012

Stochastic processes and financial applications6 references3 citations

TL;DR

This paper proposes a non-parametric, arbitrage-free model for long-term yield curve prediction using a discretized yield curve as the state variable. By embedding Heath-Jarrow-Morton (HJM) drift conditions and separating volatility estimation from risk premium calibration, the model ensures no-arbitrage while enabling accurate long-horizon forecasts, validated through backtesting and comparison with the Vasièek model.

ABSTRACT

We present an arbitrage-free non-parametric yield curve prediction model which takes the full (discretized) yield curve as state variable. We believe that absence of arbitrage is an important model feature in case of highly correlated data, as it is the case for interest rates. Furthermore, the model structure allows to separate clearly the tasks of estimating the volatility structure and of calibrating market prices of risk. The empirical part includes tests on modeling assumptions, back testing and a comparison with the Vasiček short rate model.

Motivation & Objective

• To develop a long-term yield curve prediction model that is free of arbitrage, crucial for highly correlated interest rate data.
• To model the full discretized yield curve as a stochastic process under an equivalent martingale measure.
• To separate the estimation of volatility structure from the calibration of market prices of risk for clearer model interpretation.
• To validate the model through backtesting and comparison with the Vasicek short-rate model.
• To ensure robustness by embedding no-arbitrage conditions directly into the yield curve dynamics.

Proposed method

• The model uses a discrete time grid with time step Δ, representing yield curves at regular intervals.
• Yield curve evolution is modeled via a stochastic representation involving deterministic drift, volatility, and innovation terms under the equivalent martingale measure ℙ*.
• The Heath-Jarrow-Morton (HJM) drift condition is embedded explicitly to ensure no-arbitrage in the model.
• The innovation terms ε* are ℱt-measurable and independent of ℱt−Δ under ℙ*, enabling conditional prediction.
• Volatility structure is modeled via a time- and state-dependent function vΔ, while the drift term αΔ ensures absence of arbitrage.
• Calibration is performed by estimating αΔ and vΔ from historical yield curve data, with separate treatment of risk premia and volatility.

Experimental results

Research questions

• RQ1 Can a non-parametric yield curve model be constructed that is free of arbitrage while allowing long-term predictions?
• RQ2 How can the volatility structure and market prices of risk be separated in a yield curve prediction model?
• RQ3 Does the inclusion of the HJM drift condition improve the robustness and consistency of long-term yield curve forecasts?
• RQ4 How does the model perform in backtesting compared to the Vasicek short-rate model?
• RQ5 What impact does arbitrage-free modeling have on P&L distribution shifts in risk management contexts?

Key findings

• The model ensures no-arbitrage by embedding the HJM drift condition directly into the yield curve dynamics, preventing artificial P&L shifts.
• The separation of volatility estimation and risk premium calibration allows for clearer interpretation and improved model stability.
• Backtesting confirms the model's ability to reproduce historical yield curve movements while maintaining consistency over long horizons.
• The model outperforms the Vasicek short-rate model in long-term prediction accuracy, particularly in capturing yield curve shape dynamics.
• Theoretical analysis shows that the conditional expectation of the prediction matrix converges appropriately, supporting the model's mathematical consistency.
• The use of a discrete-time grid with flexible Δ allows for practical implementation across different market frequencies (e.g., daily, weekly, monthly).