Cluster-Specific Latent Factor Estimation in High-Dimensional Financial Time Series

Stjepan Begusic,Zvonko Kostanjcar

doi:10.1109/access.2020.3021898

Abstract

Unsupervised learning methods have been increasingly used for detecting latent factors in high-dimensional time series, with many applications, especially in financial risk modelling. Most latent factor models assume that the factors are pervasive and affect all of the time series. However, some factors may affect only certain assets in financial markets, due to their clustering within countries, asset classes, or sector classifications. In this paper we consider high-dimensional financial time series with pervasive and cluster-specific latent factors, and propose a clustering and latent factor estimation method. We also develop a model selection algorithm, based on the spectral properties of asset correlation matrices and asset graphs. A simulation study with known data generating processes demonstrates that the proposed method outperforms other clustering methods and provides estimates with a high degree of accuracy. Moreover, the model selection procedure is also shown to provide stable and accurate estimates for the number of clusters and latent factors. We apply the proposed methods to datasets of asset returns from global financial markets using a backtesting approach. The results demonstrate that the clustering approach and estimated latent factors yield relevant information, improve risk modelling and reduce volatility in optimal minimum variance portfolios.

Highlights

With the rise of data driven decision making in risk management, statistical and machine learning methods are becoming increasingly important as their ability to uncover meaningful information and perform well out-of-sample is put to the test in real world scenarios
Since there is no ‘‘ground truth’’ in financial data, we develop a simulation framework based on data generating processes (DGPs) which feature heavy-tailed distributed returns and correlated residuals, in which the ground truth is known - allowing us to measure the performance of the estimation procedure and the model selection method
To estimate the number of pervasive factors and clusters, we propose a model selection method based on the spectral properties of the asset correlation matrix and the asset graph estimated from the return time series

Summary

Introduction

With the rise of data driven decision making in risk management, statistical and machine learning methods are becoming increasingly important as their ability to uncover meaningful information and perform well out-of-sample is put to the test in real world scenarios This field has recently attracted a fair amount of interdisciplinary research, bringing together mathematical, physical, econometric and computer science approaches [1]–[3]. To estimate the effects that these underlying factors have on observed asset returns, traditional modelling approaches use observable macroeconomic time series (such as GDP growth, interest rates, or market returns) as model inputs [5], while others focus on finding proxies for unobservable factors (known as size, value, or momentum) using economic firm-level data [6], [7] Recent empirical results have been challenging some of these models, giving advantage to more agnostic statistical approaches [9]

Methods

Results

Conclusion