Paper: arXiv 2309.02205
Authors: Trent Spears, Stefan Zohren, Stephen Roberts
Abstract
We consider a conditional factor model for a multivariate portfolio of United States equities in the context of analysing a statistical arbitrage trading strategy. A state space framework underlies the factor model whereby asset returns are assumed to be a noisy observation of a linear combination of factor values and latent factor risk premia. Filter and state prediction estimates for the risk premia are retrieved in an online way. Such estimates induce filtered asset returns that can be compared to measurement observations, with large deviations representing candidate mean reversion trades. Further, in that the risk premia are modelled as time-varying quantities, non-stationarity in returns is de facto captured. We study an empirical trading strategy respectful of transaction costs, and demonstrate performance over a long history of 29 years, for both a linear and a non-linear state space model. Our results show that the model is competitive relative to the results of other methods, including simple benchmarks and other cutting-edge approaches as published in the literature. Also of note, while strategy performance degradation is noticed through time – especially for the most recent years – the strategy continues to offer compelling economics, and has scope for further advancement.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematical techniques like state-space frameworks, Kalman filters, and unscented variants, indicating high complexity. It also presents a detailed empirical study with 29 years of backtesting data, transaction cost considerations, and benchmark comparisons against other methods, showing substantial empirical rigor.
Research Flowchart
flowchart TD A["Research Goal: Evaluate statistical arbitrage using a conditional factor model with latent, time-varying risk premia"] --> B["Input: 29-year history of US Equities data"] B --> C["Methodology: State Space Model (Linear & Non-linear)"] C --> D["Computational Process: Online Kalman Filter & State Prediction"] D --> E["Output: Filtered returns & Residuals (Deviations from Latent Risk Premia)"] E --> F["Trading Strategy: Mean Reversion on Residuals (Transaction Cost Adjusted)"] F --> G["Key Findings: Competitive returns over 29 years despite recent degradation; Outperforms benchmarks"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a conditional factor model embedded in a state‑space framework for statistical arbitrage trading on US equities, using Kalman and Unscented Kalman filters.
Source sentence
We consider a conditional factor model for a multivariate portfolio of United States equities in the context of analysing a statistical arbitrage trading strategy.
Assumes 6
- A 100% leverage portfolio is used, with market cost equal to the larger of long or short positions. execution
Source sentence
We assume a 100% leverage portfolio so that for each dollar invested in the portfolio, a position of up to a dollar’s worth long and / or a dollar’s worth short can be entered.
- Transaction costs are 5 basis points per trade (10 bps round‑trip). execution
Source sentence
All methods assume transaction costs of 5bps per trade.
- The S&P 500 index can be traded directly (or via futures) without considering spot/futures basis. market
Source sentence
Implicit in the hedging strategy is the assumption that we can trade long or short positions in the S&P 500 index and earn or pay the total return of the index each period.
- Risk premia are time‑varying, capturing non‑stationarity in returns. model
Source sentence
Further, in that the risk premia are modelled as time‑varying quantities, non‑stationarity in returns is de facto captured.
Show 2 more
- Linear Gaussian state‑space dynamics with known time‑varying matrices and white‑noise covariances. model
Source sentence
Here ϵxk and ϵyk model independent zero‑mean white‑noise processes with respective covariances Ψk and Ωk, assumed to be known functions of time.
- Daily data are used, selecting the top 2000 US stocks by market capitalisation. data
Source sentence
On each trading day, we select the top N greatest stocks across all available US equities sorted by market capitalisation, for N = 2000.
Data
US equities · equities · daily
- US equity daily data (CRSP and IBES) market-prices commercial
Source sentence
The raw data is daily US equity market data for the years 1992-2021 inclusive. On each trading day, we select the top N greatest stocks across all available US equities sorted by market capitalisation, for N = 2000.
CRSP I/B/E/S
Costs
Modeled
5 basis points per trade (10 bps round‑trip), applied in headline results.
Source sentence
All methods assume transaction costs of 5bps per trade.
Finds 4
- Both Kalman Filter (KF) and Unscented Kalman Filter (UKF) models dominate OLS and benchmark models on Sharpe and drawdown metrics. performance
Source sentence
Modelwise, the KF and the UKF outperform relative to OLS and BM across all but one groupwise comparisons of 1st, 2nd and 3rd quantiles, and averages, of annual Sharpe ratios.
- The non‑linear UKF model does not provide a clear outperformance over the linear KF, making its extra complexity unjustified for this application. negative
Source sentence
we did not find that the non‑linear state space model was justified relative to the linear case; though it was no worse, any outperformance benefit does not contrast well with the increased modelling complexity introduced.
- Performance degrades with higher transaction costs; at 15 bps the median annual Sharpe becomes negative. performance
Source sentence
at 15 bps of transaction costs, the median annual Sharpe ratio over the sample realises negative.
- The strategy’s drawdowns are generally milder than the market portfolio, with typical drawdowns between –5% and –7% versus market drops of –10% or worse. performance
Source sentence
our strategies realise drawdowns between -7 and -5 %, with the exception of the strategy with window size of 5 that realises a -9.5% drawdown on one of the three occasions.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 24 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.