News analytics can provide a different perspective for investors and market makers, and can help narrow the gap between the traditional way of constructing systematic models (i.e. time-based models) and the typical catalyst-driven approach applied by discretionary macro investors.
Using Sentiment to Trade Intraday Emerging Market Currencies
The importance of the sentiment feature increases with longer holding periods (from minutes to 1-2 hours). Although these longer-term signals are generally less accurate, the pattern is robust across all models. The best result is achieved for USDZAR with an accuracy of 55.4% over a 1-minute horizon.
Sentiment is Paramount in Restructuring the Predictive Framework
The potential applications of the suggested framework could be many across various asset classes and frequencies (from a few minutes which would be more relevant for market makers, to hours/days which would be most useful for the buy-side quant and discretionary firms), as well as for return or risk analysis (e.g. by predicting the impact of news arrival on realized or implied volatility).
High-Level Results of this Forex Day Trading Framework
Overall, the results are encouraging with obvious ways of further sophisticating the signals, either through more complex feature engineering or model selection or calibration.
As expected, just like in many classification problems in finance where the objective is to identify the direction of the bet rather than the size of the bet, the out-of-sample accuracy across holding periods and FX pairs is not very high. However, there are some interesting patterns in the results.Request White Paper