Testing Autoregressive Models for Short-Term FX Trading

Summary

The article explains how an autoregressive model predicts the next exchange-rate value from prior observations, then examines whether those predictions could support AUD/USD trades. It discusses partial autocorrelation across several sampling intervals, fits an AR(10) model on historical prices, and describes applying its fixed coefficients to later data in a simulated strategy. The analysis finds little basis for treating the most recent minutely prices as a distinctive signal and cautions that the approach may not be tradable under retail spot FX conditions.

The article suggests possible refinements: trade only when forecasts differ sufficiently from the current price, avoid low-volatility periods, and periodically refit the model. It recommends checking whether coefficients persist across successive estimation windows. These are proposed avenues rather than demonstrated improvements: the document supplies no strategy performance results, and it emphasizes that trading costs and weak predictive value constrain practical use.

Key ideas

• An autoregressive model uses past values of a series to forecast its next observation.
• Partial autocorrelation can help examine relationships between prices at different lags and sampling intervals.
• The article applies fixed AR(10) coefficients to later AUD/USD data as a trading experiment.
• Forecast thresholds, volatility filters, and periodic refitting are proposed for further investigation.
• The discussion cautions that short-term FX predictability may not overcome retail trading costs.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.