The Cross-Quantilogram: Measuring Quantile Dependence and Testing Directional Predictability between Time Series

Heejoon Han, Oliver Linton, Tatsushi Oka, Yoon-Jae Whang

TL;DR

This paper introduces the "cross-quantilogram," a new tool to measure and test for quantile dependence between two time series. This allows for a more nuanced understanding of how one series might predict another, especially in the tails of their distributions, and is applied to analyze stock return predictability and systemic risk.

What Problem Does This Paper Address?

Traditional methods for analyzing the relationship between time series, like the cross-correlation function, often focus on the mean relationship and assume certain moment conditions (like finite variance). This can miss important dynamics, especially in financial data which often exhibits heavy tails and extreme events. Existing methods for testing directional predictability, such as the quantilogram, have limitations in their theoretical guarantees for confidence intervals and their ability to handle general dependence structures. This paper aims to fill this gap by providing a robust and flexible tool for analyzing quantile dependence and testing for directional predictability across various parts of the distribution.

Methods

The core contribution is the cross-quantilogram, which extends the concept of the quantilogram to analyze the dependence between two time series. It measures the serial dependence between "quantile hits" – events where a time series falls below a specific conditional quantile. The paper establishes the asymptotic distribution of the cross-quantilogram and develops a test statistic for directional predictability. To address the issue of nuisance parameters in the limiting distributions, they propose and prove the validity of a stationary bootstrap procedure for constructing confidence intervals. Additionally, they introduce a self-normalized version of the test statistic, which has an asymptotically pivotal distribution, simplifying inference by avoiding the need for long-run variance estimation or bootstrapping. The paper also extends these methods to a multivariate setting and allows for the use of conditional quantiles, enabling control for other information.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and simulation studies, with empirical applications using publicly available financial data.

Key Insights and Ideas

  • Beyond the Mean: The cross-quantilogram reveals relationships that linear models or traditional cross-correlations might miss, particularly in extreme quantiles.
  • Robustness to Heavy Tails: By relying on quantile hits, the method is inherently robust to heavy-tailed distributions, a common feature in financial time series.
  • Directional Predictability: The method specifically tests for "directional predictability," meaning whether knowing the past of one series helps predict whether the other series will be above or below a certain quantile.
  • Bootstrap for Inference: The stationary bootstrap provides a practical way to construct valid confidence intervals, even with complex dependence structures.
  • Self-Normalization Simplifies Inference: The self-normalized statistic offers a pivotal distribution, making critical value determination more straightforward.

Key Results

The paper provides theoretical results establishing the asymptotic distributions of the cross-quantilogram and related test statistics under general weak dependence conditions (strong mixing). They demonstrate the validity of their stationary bootstrap procedure for constructing confidence intervals and show that the self-normalized statistic is asymptotically pivotal. Simulation studies suggest that the proposed methods have good size and power performance in finite samples. Empirical applications to stock return predictability and systemic risk reveal interesting patterns of quantile dependence, showing how risk (stock variance) can predict stock returns at different quantiles and how individual financial institutions' distress relates to market-wide distress.

Limitations and Future Work

The authors acknowledge that the theoretical framework is developed for strictly stationary and strong mixing sequences. While they apply it to financial data that might exhibit non-stationarity (e.g., during the financial crisis), they note that a rigorous treatment of non-stationary time series in this context is a challenging issue and will be reported in future work. They also mention that extending the self-normalized approach to a range of quantiles is not straightforward. Future work could explore these extensions and further investigate the application of the cross-quantilogram to other economic and financial phenomena.

Summary updated

Abstract

This paper proposes the cross-quantilogram to measure the quantile dependence between two time series. We apply it to test the hypothesis that one time series has no directional predictability to another time series. We establish the asymptotic distribution of the cross-quantilogram and the corresponding test statistic. The limiting distributions depend on nuisance parameters. To construct consistent confidence intervals we employ a stationary bootstrap procedure; we establish consistency of this bootstrap. Also, we consider a self-normalized approach, which yields an asymptotically pivotal statistic under the null hypothesis of no predictability. We provide simulation studies and two empirical applications. First, we use the cross-quantilogram to detect predictability from stock variance to excess stock return. Compared to existing tools used in the literature of stock return predictability, our method provides a more complete relationship between a predictor and stock return. Second, we investigate the systemic risk of individual financial institutions, such as JP Morgan Chase, Morgan Stanley and AIG. Journal of Econometrics, 193(1), 251-270.
BibTeX
@article{han2016cross,
  author = {Heejoon Han and Oliver Linton and Tatsushi Oka and Yoon-Jae Whang},
  title = {{The Cross-Quantilogram: Measuring Quantile Dependence and Testing Directional Predictability between Time Series}},
  journal = {Journal of Econometrics},
  year = {2016},
  doi = {10.1016/j.jeconom.2016.03.001},
}
Figure