Option Pricing and Hedging for Discrete Time Autoregressive Hidden Markov Model

Massimo Caccia, Bruno N. RémillardarXiv 2017 · OpenReview

Abstract

In this paper, we solve the discrete-time mean-variance hedging problem when asset returns follow a multivariate autoregressive hidden Markov model (ARHMM). Time-dependent volatility and serial dependence are well-established properties of financial time series, and our model accounts for both. To illustrate the relevance of our proposed methodology, we compare the ARHMM with the standard hidden Markov model (HMM) via likelihood ratio tests and a novel goodness-of-fit test on S&P 500 daily returns. Furthermore, we present out-of-sample hedging results on S&P 500 vanilla options and a trading strategy based on theoretical prices, comparing it to simpler models, including the classical Black-Scholes delta-hedging approach.

Introduction

The search for a "perfect" option pricing model has been a central theme in finance since the seminal work of Black and Scholes (1973). However, the classical Black-Scholes framework relies on two significant simplifications: constant volatility and serial independence of returns (the "random walk" hypothesis). In practice, financial markets are characterized by "volatility clustering"—where high-volatility periods follow high-volatility periods—and short-term trends or mean-reversion.

While standard Hidden Markov Models (HMMs) can capture shifting volatility regimes (e.g., distinguishing between "bull" and "bear" markets), they often fail to account for the serial dependence of returns within those regimes. This paper addresses this gap by introducing a tractable model that integrates both time-varying volatility and autocorrelation to minimize the quadratic error of replication, a process known as optimal mean-variance hedging.

The Autoregressive Hidden Markov Model (ARHMM)

The proposed ARHMM generalizes the standard regime-switching model by incorporating an autoregressive parameter ($\Phi_j$) for each state. In this framework, returns follow a Gaussian AR(1) process where the mean, volatility, and autoregressive parameters are all contingent on a hidden Markovian regime.

Parameter Estimation and Filtering

The authors utilize the Expectation-Maximization (EM) algorithm, specifically adapted for the ARHMM, to identify hidden structures in historical data. Because regimes are unobservable, a filtering algorithm is employed to determine the "most probable regime" at any given time based on the history of observed prices. To ensure model validity, the authors introduce a novel goodness-of-fit test based on the Rosenblatt transform and parametric bootstrap to determine the optimal number of regimes for a given dataset.

Computational Efficiency

To overcome the limitations of computationally intensive Monte Carlo simulations, the authors derive semi-exact techniques using piecewise linear approximations. These methods allow for the efficient calculation of expectations required for optimal hedging, significantly speeding up the computation of Greeks and hedging weights for real-time applications.

Empirical Methodology and Data

The model's performance was evaluated using daily log-returns of the S&P 500 Total Return index. The analysis focused on three distinct historical windows to test the model under varying market conditions: the 2000s Recovery (2004-2008), the 2008-2009 Financial Crisis, and the 2010s Recovery (2010-2017).

The stationarity of these periods was verified using a nonparametric changepoint test. The resulting P-values, shown in the table below, support the selection of these windows as statistically distinct market environments.

P-values for the nonparametric change point test across different market periods.

Hedging Results and Discussion

The inclusion of an autoregressive component proves transformative. It allows the model to capture mean-reversion, a well-established "stylized fact" of financial series that standard HMMs often obscure. The ARHMM identifies that different regimes exhibit distinct levels of mean-reversion; for example, during the 2000s recovery, strong mean-reversion was observed in both low-volatility "bull" and high-volatility "correction" regimes.

The model effectively "paints" the market into these volatility states. By mapping the most probable regime onto the cumulative performance of the S&P 500, we can observe how high-volatility states (represented by darker vertical bands in the figure below) often coincide with market dips or periods of stagnation.

Cumulative performance of the S&P 500 (2004-2008) with shaded regions representing the most probable regimes identified by the ARHMM.

Key Performance Metrics

  • Error Reduction: In simulations, the Optimal Hedging ARHMM (OH-ARHMM) achieved a 33% reduction in Root Mean Square Error (RMSE) compared to the Black-Scholes model and a 26% reduction compared to a standard HMM.
  • Profitability: In out-of-sample backtesting on S&P 500 vanilla options, the ARHMM-based trading strategy was the most profitable in five out of eight tested scenarios.
  • Crisis Resilience: During the 2008 Financial Crisis, the ARHMM demonstrated superior capability in isolating market downturns and adjusting hedging weights accordingly, whereas simpler models incurred substantial losses.

Conclusion

The ARHMM framework provides a significant advancement in option pricing and hedging by simultaneously addressing volatility clustering and serial dependence. While the authors acknowledge potential biases in using the "most probable regime" for hedging, the statistical and economic gains are substantial. Future research may extend this framework to multivariate time series for portfolio hedging and apply it to American options with early-exercise features.

Summary updated

Abstract

In this paper we solve the discrete time mean-variance hedging problem when asset returns follow a multivariate autoregressive hidden Markov model. Time dependent volatility and serial dependence are well established properties of financial time series and our model covers both. To illustrate the relevance of our proposed methodology, we first compare the proposed model with the well-known hidden Markov model via likelihood ratio tests and a novel goodness-of-fit test on the S\&P 500 daily returns. Secondly, we present out-of-sample hedging results on S\&P 500 vanilla options as well as a trading strategy based on theoretical prices, which we compare to simpler models including the classical Black-Scholes delta-hedging approach.
BibTeX
@misc{caccia2017option,
  author = {Massimo Caccia and Bruno N. Rémillard},
  title = {{Option Pricing and Hedging for Discrete Time Autoregressive Hidden Markov Model}},
  year = {2017},
  eprint = {1707.02019},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1707.02019},
}
Figure