FX traders have long used classical strategies such as the carry trade, trend-following, and value dynamics, to trade the foreign exchange markets, but how much can sentiment, aggregated from hundreds of thousands of financial news articles, help enhance these traditional factors?
This was the question posed by RavenPack Data Scientists in their latest study.
The first step was to see how well traditional factors worked on their own, in order to have a benchmark for comparison.
They then ran a similar backtest but with the same factors enhanced by RavenPack news sentiment.
Hurdles and Solutions
The RavenPack platform’s versatile suite of sentiment indicators enabled researchers to get around certain obstacles that arose.
How news about interest rates should be interpreted from a sentiment perspective, was one such hurdle, for example.
The conventional economic interpretation is that raising interest rates is negative for the economy, as it implies rising inflation, yet from the standpoint of a nation’s currency it is mostly positive.
In this case, RavenPack’s main ESS sentiment indicator was prone to misread the news. The ESS is trained on expert opinion and a more conventional economic interpretation.
Researchers got around the problem by using the platform’s more flexible, language-based CSS sentiment gauge, enabling a more nuanced read of the complex interpretations around interest rate news and its impact on FX.
The chart below compares traditional factors to those enhanced with sentiment and shows how the addition of sentiment helped optimize returns in all 3 cases.
When it came to which of the sentiment metrics - ESS or CSS - was more effective, the chart below shows that CSS outperformed in all 3 cases, not just for the more interest rate sensitive carry trade.
Overall, a composite portfolio that invested equally in all 3 factors enhanced by news sentiment, delivered an Annualized Return of 210bps with an information ratio (IR) of 0.70.
This was over 3x better risk-adjusted returns than the benchmark’s strategy. Moreover, hedging the sentiment-enhanced portfolio by shorting the benchmark, yielded an Annualized Return of 156bps with an IR of 1.05.
“This study clearly shows how sentiment can be a valuable overlay to an existing Forex multi-factor model by providing a dynamic weighing mechanism over traditionally slower moving signals. Sentiment consistently adds value over time and works well with both long and short currency baskets,” says Peter Hafez, Chief Data Scientist at RavenPack.Request White Paper