Forecasting USDJPY with Bond Data and a Tuned Support Vector Regressor

Summary

The article explores whether US and Japanese 10-year government bond data can help forecast USDJPY. It compares three predictor sets: currency OHLCV alone, bond OHLCV, and a combined set. The target is a future USDJPY close, and the workflow includes joining minute-level series, examining charts and correlations, and evaluating models with time-series cross-validation that preserves order. Although bond prices were strongly negatively correlated with USDJPY in the sample, the article reports that currency-only predictors achieved the lowest test error.

Linear regression provided the benchmark, while a tuned linear support vector regressor was selected as the candidate and reportedly exceeded that benchmark on validation data. The model was exported for use in a MetaTrader expert advisor that compares forecasts with current prices and manages positions. The reported relationships and model comparisons are specific to the chosen data and setup; the document does not establish that bond inputs improve forecasts or that the example trading logic will remain profitable out of sample.

Key ideas

• The study compares currency-only, bond-only, and combined inputs for a future USDJPY close forecast.
• The historical bond series showed strong negative correlation with USDJPY, but currency-only inputs produced the lowest test error.
• Time-series cross-validation was used without random shuffling.
• A tuned linear support vector regressor reportedly beat the linear regression benchmark on validation data.
• The exported model was integrated into an expert advisor, but the reported results do not establish durable trading profitability.

Tags

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