By Saeed Amen, founder of Cuemacro.
I recently wrote a paper looking at how equity traders can use machine-readable FT news articles to create systematic trading strategies for large-cap US stocks.
The rationale behind using the FT as a source for news, is that the FT publishes a lot of market-moving news. Furthermore, the news published by the FT is likely to be complementary to other news sources, and it covers the most important market events. Indeed, this is why discretionary traders have been reading the FT for many years.
The question we want to ask is the following: can we replicate the way a discretionary trader reads the FT, to harvest alpha from the FT when it is in machine-readable form? The motivating factor for doing this, is that a computer can parse much more news than a human would ever have time to read.
Creating a news sentiment indicator and interpreting it
The approach that was used in the paper was to filter FT news articles for each equity ticker in our trading universe. A news sentiment score was then calculated for each of these articles. Later, news sentiment indicators were created out of these news sentiment scores at both daily and intraday time horizons.
In a stylised example below, we can see the 2 week future returns of TSLA stock versus the daily sentiment indicator for TSLA news in the FT. Most of the subsequent positive returns occur when news sentiment for TSLA articles is positive. It’s noted in the paper that in the majority of stocks in our trading universe of large-cap US equities, the daily sentiment indicator had a positive correlation with future results.
By contrast, in this next stylised example showing TSLA's stock price against TSLA intraday sentiment, we see that most of the stock price rally occurred during periods of negative sentiment. In other words, it suggests that at intraday time horizons, we should try to fade sentiment.
Backtesting systematic trading strategies for large-cap US equities using FT news
A number of different rules were then backtested both for daily and intraday sentiment for large-cap US stocks, which either went long a stock or flat depending on the underlying sentiment.
It was found that on a daily basis, the trading rules that followed the trend in sentiment were more profitable than those which used mean reversion of sentiment. By contrast, on an intraday basis, buying stocks with negative intraday sentiment was more profitable than buying stocks with positive intraday sentiment.
This suggested that the market might overreact on an intraday basis to news, whilst at daily/weekly time horizons positive sentiment persisted more, which tallies with the previous observations above.
Lastly, a portfolio of our best performing daily and intraday rules was created. This portfolio was leveraged up by two times, so its volatility was comparable with that of S&P 500, which was our benchmark. The next chart presents the returns of both our sentiment portfolio and S&P 500.
Our news sentiment portfolio had risk-adjusted returns of 1.0, outperforming the S&P 500 over the same time horizon, which had risk-adjusted returns of 0.8. Furthermore, our news sentiment portfolio had higher returns and smaller drawdowns than our benchmarks. The improvement both in risk-adjusted returns and drawdowns suggests that this news dataset adds value compared to a long-only benchmark.
Want to read more about using machine-readable FT news to trade single stocks?
Click below to download the full paper on using FT news to trade large-cap US equities.