Graph Learning for Foreign Exchange Rate Prediction and Statistical Arbitrage
The paper introduces a two step graph approach to FX prediction and statistical arbitrage that encodes multicurrency links, interest rate parity, and execution time risk. FX rates are treated as edge regression on a graph. The arbitrage module maximizes information ratio under lag and enforces no arbitrage exactly via projection and ReLU. Using 1995 to 2024 data, GNN beats MLPs. The strategy raises information ratio 61.89 percent and Sortino 45.51 percent, with lower volatility, drawdown.
What it examines
The paper proposes a two-step graph learning framework for FX. Step 1 predicts exchange rates via edge-level regression on a spatiotemporal currency--interest-rate graph, with MLE-based currency values. Step 2 optimizes statistical arbitrage under execution lags, enforcing constraints by projection/ReLU and maximizing risk‑adjusted return with proven feasibility.
What it concludes
Experiments on ten major currencies show significant FXRP MSE gains and FXSA risk-adjusted outperformance (Information Ratio +61.89%, Sortino +45.51%) with lower volatility and drawdowns. Applications: FX trading, market making, hedging overlays, risk control. Limits: daily data, limited universe, costs ignored. Future: intraday, transaction costs, richer graphs.
Evidence objects
Researchers debut a two-step graph method: currencies nodes, exchanges edges; IRs node features, FX edge features; prediction with MLE-derived currency values, plus FX StatArb maximizing ratio, arbitrage enforced by projection/ReLU.
key_findings bullet 1 · key_findings · validation V0
Using Finaeon data (1995--2024) on 10 currencies and 1/2/5/10-year rates, the GNN tops MLPs in MSE; FXSA lifts ratio 61.89%, Sortino 45.51%, cuts volatility 52.23%, drawdown 44.77%, trimming returns 22.73%.
key_findings bullet 2 · key_findings · validation V0
Notably, the approach provably satisfies arbitrage constraints, outperforming penalty relaxations. Strengths: principled enforcement, lag-aware design, rigorous out-of-sample validation. Limitations: daily-close granularity, simplified borrowing/transactions, forward-filling, excluding CNY, evaluating only 10 currencies.
key_findings bullet 3 · key_findings · validation V0
Introduces a two-step FX framework: edge-level spatiotemporal graph regression using interest-rate and MLE currency-value features, followed by stochastic-optimization statistical arbitrage modeling observation--execution lags with projection/ReLU ensuring provable constraints and an exchange influence graph. Demonstrates significant MSE and risk-adjusted gains. Novel integration within FX is distinctive and practical, though not unprecedented.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
Abstract: We propose a two-step graph learning approach for foreign exchange statistical arbitrages (FXSAs), addressing two key gaps in prior studies: the absence of graph-learning methods for foreign exchange rate prediction (FXRP) that leverage multi-currency and currency-interest rate relationships, and the disregard of the time lag between price observation and trade execution. In the first step, to cap… ▽ More We propose a two-step graph learning approach for foreign exchange statistical arbitrages (FXSAs), addressing two key gaps in prior studies: the absence of graph-learning methods for foreign exchange rate prediction (FXRP) that leverage multi-currency and currency-interest rate relationships, and the disregard of the time lag between price observation and trade execution. In the first step, to capture complex multi-currency and currency-interest rate relationships, we formulate FXRP as an edge-level regression problem on a discrete-time spatiotemporal graph. This graph consists of currencies as nodes and exchanges as edges, with interest rates and foreign exchange rates serving as node and edge features, respectively. We then introduce a graph-learning method that leverages the spatiotemporal graph to address the FXRP problem. In the second step, we present a stochastic optimization problem to exploit FXSAs while accounting for the observation-execution time lag. To address this problem, we propose a graph-learning method that enforces constraints through projection and ReLU, maximizes risk-adjusted return by leveraging a graph with exchanges as nodes and influence relationships as edges, and utilizes the predictions from the FXRP method for the constraint parameters and node features. Moreover, we prove that our FXSA method satisfies empirical arbitrage constraints. The experimental results demonstrate that our FXRP method yields statistically significant improvements in mean squared error, and that the FXSA method achieves a 61.89% higher information ratio and a 45.51% higher Sortino ratio than a benchmark. Our approach provides a novel perspective on FXRP and FXSA within the context of graph learning. △ Less
Source row: 1001 · abstract type: unknown