Statistical Arbitrage via Single-View and Multi-View Spectral Clustering on Mixed Frequency Data

Abstract

Systematically identifying clusters of similar assets is a critical step in statistical arbitrage strategies. This paper makes two key contributions: (i) In addition to assets' daily closing prices, we incorporate realized estimators (i.e. realized volatility, realized beta, and others) derived from high-frequency intraday data into the daily feature set; and (ii) We examine the effectiveness of single-view and multi-view spectral clustering algorithms on these feature sets for identifying similar assets. We use these clustered assets to construct a variety of signals and trading rules, including the distance method, cointegration method, copula method, and an optimal entry-exit rule based on Ornstein-Uhlenbeck spread dynamics. We evaluate our methodology on the S&P 500 equities, futures contracts, exchange traded funds, and foreign exchange asset classes. Our findings show that trading profitability is influenced more by the selection of feature sets and clustering methods than by the choice of signals or trading rules. Specifically, strategies employing multi-view clustering algorithms, which integrate various realized estimators, consistently deliver superior risk-adjusted returns and reduced drawdowns compared to those based on the distance method and single-view clustering that rely solely on daily closing price data.

Keywords: statistical arbitrage, pairs trading, high-frequency data, realized estimators, machine learning algorithms

JEL Classification: C45, C58, G12