Forecasting Forex OHLC Changes with Vector Autoregression

Summary

The article explains vector autoregression as a multivariate forecasting method in which each series depends on its own lagged values and the lags of other series. It applies VAR to daily EURUSD open, high, low, and close data: the example differences the price series to seek stationarity, checks them with the Augmented Dickey-Fuller test, reviews correlations, and compares information criteria to choose a lag order. It then describes producing out-of-sample OHLC forecasts and using predicted price relationships to form trading signals.

The example reports low stationarity-test p-values for the differenced series and presents lag-selection results, but these diagnostics do not establish trading profitability. The article notes that VAR assumes linear relationships, stationary inputs, adequate observations, no perfect multicollinearity, and uncorrelated residuals. It also cautions that financial series may violate these assumptions and that using many variables or lags can overfit. The method is presented as an educational trading application, not as evidence of a validated strategy.

Key ideas

• VAR models each series using lagged values from all variables in the system.
• The example transforms EURUSD OHLC levels by differencing them before fitting VAR.
• Stationarity tests, correlation checks, and information criteria are used to assess inputs and select lags.
• The article describes translating out-of-sample forecasts into trading signals.
• VAR's linearity and stationarity assumptions, along with overfitting risk, limit its reliability in markets.

Tags

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