Why do exchange rates often move in ways that even the best models can’t predict? For decades, researchers have found that “random-walk” forecasts can outperform models based on fundamentals (Meese & Rogoff, 1983a; Meese & Rogoff, 1983b). That’s puzzling. Theory says fundamental variables should matter. But in practice, FX markets react so quickly to new information that they often seem unpredictable (Fama, 1970; Mark, 1995).
Why Traditional Models Fall Short
To get ahead of these fast-moving markets, later research looked at high-frequency, market-based signals that move ahead of big currency swings. Spikes in exchange‐rate volatility and interest‐rate spreads tend to show up before major stresses in currency markets (Babecký et al., 2014; Joy et al., 2017; Tölö, 2019). Traders and policymakers also watch credit‐default swap spreads for sovereign debt, since widening spreads signal growing fears about a country’s ability to meet its obligations. At the same time, global risk gauges, like the VIX index, which measures stock‐market volatility expectations, often warn of broader market jitters that can spill over into foreign‐exchange markets.
In recent years, machine learning has taken FX forecasting a step further. These models combine many inputs like liquidity metrics, option-implied volatility, credit spreads, and risk indexes into early-warning systems.
Tools like random forests, gradient boosting, and neural networks can detect complex, non-linear patterns that traditional models miss (Casabianca et al., 2019; Tölö, 2019; Fouliard et al., 2019).
But even these advanced models often depend on fixed-lag indicators — data points taken at specific intervals in the past, like yesterday’s interest-rate spread or last week’s CDS level. These snapshots may miss how stress gradually builds or unfolds across time. In other words, they often ignore the path the data took to get there.
From Snapshots to Shape: A Better Way to Read Market Stress
A promising shift is to focus not just on past values, but on the shape of how those values evolved. This is where path-signature methods come in. Drawn from rough-path theory, these tools turn a sequence of returns into a kind of mathematical fingerprint — one that captures the twists, and turns of market movements.
Early studies show that these shape-based features can improve forecasts for both volatility and FX forecasts, offering a more dynamic view of market behavior.
What This Means for Forecasting and Risk Management
These findings suggest that the path itself — how returns unfold over time — can to predict asset price movements and market stress. By analyzing the full trajectory of recent returns rather than isolated snapshots, analysts can detect subtle shifts in market behavior that predicts moves.
For anyone managing currency risk — central banks, fund managers, and corporate treasury teams — adding these signature features to their toolkit may offer earlier and more reliable warnings of FX trouble—giving decision-makers a crucial edge.
Looking ahead, path-signature methods could be combined with advanced machine learning techniques like neural networks to capture even richer patterns in financial data.
Bringing in additional inputs, such as option-implied metrics or CDS spreads directly into the path-based framework could sharpen forecasts even more.
In short, embracing the shape of financial paths — not just their endpoints — opens new possibilities for better forecasting and smarter risk management.
References
Babecký, J., Havránek, T., Matějů, J., Rusnák, M., Šmídková, K., & Vašíček, B. (2014). Banking, Debt, and Currency Crises in Developed Countries: Stylized Facts and Early Warning Indicators. Journal of Financial Stability, 15, 1–17.
Casabianca, E. J., Catalano, M., Forni, L., Giarda, E., & Passeri, S. (2019). An Early Warning System for Banking Crises: From Regression‐Based Analysis to Machine Learning Techniques. Dipartimento di Scienze Economiche “Marco Fanno” Technical Report.
Cerchiello, P., Nicola, G., Rönnqvist, S., & Sarlin, P. (2022). Assessing Banks’ Distress Using News and Regular Financial Data. Frontiers in Artificial Intelligence, 5, 871863.
Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work. Journal of Finance, 25(2), 383–417.
Fouliard, J., Howell, M., & Rey, H. (2019). Answering the Queen: Machine Learning and Financial Crises. Working Paper.
Joy, M., Rusnák, M., Šmídková, K., & Vašíček, B. (2017). Banking and Currency Crises: Differential Diagnostics for Developed Countries. International Journal of Finance & Economics, 22(1), 44–69.
Mark, N. C. (1995). Exchange Rates and Fundamentals: Evidence on Long‐Horizon Predictability. American Economic Review, 85(1), 201–218.
Meese, R. A., & Rogoff, K. (1983a). The Out‐of‐Sample Failure of Empirical Exchange Rate Models: Sampling Error or Misspecification? In J. A. Frenkel (Ed.), Exchange Rates and International Macroeconomics (pp. 67–112). University of Chicago Press.
Meese, R. A., & Rogoff, K. (1983b). Empirical Exchange Rate Models of the Seventies. Journal of International Economics, 14(1–2), 3–24.
Tölö, E. (2019). Predicting Systemic Financial Crises with Recurrent Neural Networks. Bank of Finland Technical Report.
