Paper: arXiv 2512.02037

Authors: Marek Adamczyk, Michał Dąbrowski

Abstract

We study a systematic approach to a popular Statistical Arbitrage technique: Pairs Trading. Instead of relying on two highly correlated assets, we replace the second asset with a replication of the first using risk factor representations. These factors are obtained through Principal Components Analysis (PCA), exchange traded funds (ETFs), and, as our main contribution, Long Short Term Memory networks (LSTMs). Residuals between the main asset and its replication are examined for mean reversion properties, and trading signals are generated for sufficiently fast mean reverting portfolios. Beyond introducing a deep learning based replication method, we adapt the framework of Avellaneda and Lee (2008) to the Polish market. Accordingly, components of WIG20, mWIG40, and selected sector indices replace the original S&P500 universe, and market parameters such as the risk free rate and transaction costs are updated to reflect local conditions. We outline the full strategy pipeline: risk factor construction, residual modeling via the Ornstein Uhlenbeck process, and signal generation. Each replication technique is described together with its practical implementation. Strategy performance is evaluated over two periods: 2017-2019 and the recessive year 2020. All methods yield profits in 2017-2019, with PCA achieving roughly 20 percent cumulative return and an annualized Sharpe ratio of up to 2.63. Despite multiple adaptations, our conclusions remain consistent with those of the original paper. During the COVID-19 recession, only the ETF based approach remains profitable (about 5 percent annual return), while PCA and LSTM methods underperform. LSTM results, although negative, are promising and indicate potential for future optimization.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced statistical methods (PCA, Ornstein-Uhlenbeck process, LSTM networks) and derives trading signals via sophisticated modeling, but also demonstrates backtest-ready implementation with detailed performance metrics (Sharpe ratios, cumulative returns) across two distinct market periods, and addresses real-world constraints like transaction costs.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Statistical Arbitrage in Polish Equities<br>Using Deep Learning Techniques"] --> B["Data Inputs<br>WIG20, mWIG40, Sector Indices<br>2017-2020"]
  B --> C["Methodology: Replication Techniques<br>PCA | ETFs | LSTM"]
  C --> D["Computational Process<br>Residual Modeling via Ornstein-Uhlenbeck<br>Signal Generation"]
  D --> E["Key Findings: 2017-2019<br>PCA: 20% Return, Sharpe 2.63<br>All methods profitable"]
  D --> F["Key Findings: 2020 Recession<br>ETFs: 5% Return<br>PCA & LSTM underperform<br>LSTM shows future potential"]

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 introduces a deep‑learning based replication method using Long Short‑Term Memory networks for statistical arbitrage pairs trading in the Polish market.

Source sentence
as a new method that was not covered in the literature of the subject yet: considering Long short-term memory networks (LSTMs).

Assumes 5

  • Stock price process is continuous and returns are normally distributed. model
    Source sentence
    For simplicity we are going to assume that such process is continuous. Then, for a time interval dt (f.e. 1/252 representing
  • Linear regression with constant beta decomposes returns into drift, market factor and idiosyncratic noise. model
    Source sentence
    which is a simple regression model (commonly called 2-parameters model) decomposing stock’s return into deterministic drift αdt, market (systematic) βFt factor and a stock-unique, mean 0, uncorrelated (idiosyncratic) process ϵt
  • Residuals follow an Ornstein‑Uhlenbeck mean‑reverting process. statistical
    Source sentence
    ... Ornstein–Uhlenbeck process is going to be our natural candidate fulfilling all the requirements
  • Transaction cost of 0.1% per buy/sell is applied to all trades. execution
    Source sentence
    Transaction cost of c = 0.1% per buy/sell is going to be added.
Show 1 more
  • Backtesting uses 60 most influential companies from WIG20 and mWIG40. other
    Source sentence
    Applying techniques of Statistical Arbitrage presented by Avellaneda and Lee [4] to polish stock exchange required switching from around 500 stocks’ portfolios to just 60 most influential companies gathered by two main indices: WIG20 and mWIG40.

Data

Backtest sample: 2017 – 2019 · Polish equities (WIG20, mWIG40) · equities · daily

  • Polish equity constituents (WIG20 and mWIG40) · Warsaw Stock Exchange market-prices public
    2017-2020 (backtesting) and 2019-2021 (L · 60 most influential Polish companies
    Source sentence
    ... to just 60 most influential companies gathered by two main indices: WIG20 and mWIG40.
  • Sector indices (sWIG80 etc.) · Warsaw Stock Exchange market-prices public
    sector‑specific Polish stocks
    Source sentence
    indices are way easier to replicate than WIG because of smaller components’ number.
  • ETF based replication factors alternative unknown
    Polish market ETFs
    Source sentence
    ... considering Long short-term memory networks (LSTMs).
  • ING Bank Śląski stock data market-prices unknown
    2019-2021 (training) and 2022 (testing) · single Polish bank stock
    Source sentence
    Let us now analyse how the stacked LSTM is working on real data- in the first example ING Bank Polski

Costs

Modeled Costs are part of the reported headline results.

c = 0.1% per buy/sell (both opening and closing trades)

Source sentence
Transaction cost of c = 0.1% per buy/sell is going to be added.

Finds 4

Reported performance (PCA replication pairs trading): Sharpe 2.63 · total return 20.0%

  • PCA replication yields roughly 20% cumulative return and an annualized Sharpe ratio up to 2.63 over 2017‑2019. performance
    Source sentence
    All methods yield profits in 2017-2019, with PCA achieving roughly 20 percent cumulative return and an annualized Sharpe ratio of up to 2.63.
  • During the 2020 COVID‑19 recession only the ETF‑based approach remains profitable (~5% annual return), while PCA and LSTM underperform. performance
    Source sentence
    During the COVID-19 recession, only the ETF based approach remains profitable (about 5 percent annual return), while PCA and LSTM methods underperform.
  • LSTM overall portfolio Sharpe ratios are 0.60 (2017), 2.09 (2018) and –1.53 (2019), with a final return around 10%, worse than PCA. performance
    Source sentence
    ... The final return is around 10% which is worse than with both PCA approaches.
  • LSTM sector‑specific Sharpe ratios vary, with highest values in 2018 and many negative ratios in 2019. performance
    Source sentence
    LSTM is underperforming in 2019 for most of the groups. Highest ratios were achieved in 2018 with 2017 giving more than half of positive ratios.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 41 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.