| March 10, 2020
Advances in technology now mean computers can rapidly scan vast quantities of online news quicker than the blink of an eye. Their subtle artificial intelligence and language processing capabilities also make it possible for them to distinguish authorial viewpoints and measure Forex ‘sentiment’.
Such systems are then used by investors to drive profitable news-based trading strategies. In this article, we will look at their application in the FX space.
The Forex market experiences daily multi-trillion-dollar turnover and is known for being highly efficient at discounting novel information - one of the reasons why people say it is so difficult to trade. Yet research suggests that even in such an efficient market, AI systems, that are able to automate news reading and gauge sentiment, can still provide traders with a significant edge.
In an early study conducted by data scientists here at RavenPack, signals derived from Eurozone and U.S. economic news sentiment, were shown to provide a robust basis for a profitable
EUR/USD trading strategy
The study showed that EUR/USD actually ‘overreacted’ to positive or negative macro news over a 24-hour timeframe and that this characteristic could be used as the basis for a mean reversion or ‘contrarian’ strategy, in which the trader entered a position in the opposite direction to that implied by the news.
Interestingly, the study found that on a monthly basis, the opposite happened - the pair tended to underreact, suggesting a momentum strategy might be a more profitable fit in the medium-term.
The results from the daily strategy were positive: “A contrarian strategy with a one-day holding period on EURUSD, generates an average return of 7.2% per year with an Information Ratio of 0.75 after transaction costs,” says Peter Hafez, Chief Data Scientist at RavenPack.
More recently a study by a major U.S. investment bank found that the same contrarian effect in evidence when it came to exchange rate volatility, with major G10 FX pairs generating alpha when RavenPack news sentiment was used as the basis for a contrarian option trading strategy.
Another study found a
news sentiment-based model
correctly predicted the direction of the South African Rand - which is considered difficult to forecast even by FX standards - 55.4% of the time, from a sample of almost 100k trades, using a one-minute holding period. Considering the large sample the result is highly significant.
The short holding period of the ZAR strategy reflects the popularity of this type of data amongst high-frequency trading (HFT) hedge funds or ‘shops’ although that doesn’t mean it cannot be applied to longer-term discretionary investing too.
It is not just in forecasting FX rates that AI-generated news sentiment data can be applied - there is also a substantial amount of research showing it can be employed as an indicator of common FX drivers, such as central bank policy changes and fluctuations in
A study of how automated news text analysis could be applied to establishing a consensus estimate for Bank of Indonesia (BOI) policy rates, conducted by researchers at the
Bank of International Settlements (BIS)
, showed how a machine news-based model predicted the eventual BOI decision with unerring accuracy and was better at it than an industry-standard consensus estimate, based on Bloomberg’s survey of economists, as shown in the table of results below.
Indeed, the low frequency of many macroeconomic news releases, such as GDP, which is only published at quarterly intervals, also lends itself to forecasting using AI-generated news sentiment-based indicators, since the long time-lapse between each official release provides a space in which news sentiment for factors that will impact GDP can be used as the basis for generating an early estimate.
A recent study by Copenhagen Economics, for example, used the RavenPack analytics platform to show how AI aggregated news sentiment for news about the Chinese economy, provided a statistically sound basis for the forecasting of major macro-economic releases such as GDP, balance-of-payments and unemployment.
“Including ESS indexes improves the forecasts of macroeconomic time series for China,” says Asger Lunde, Director, Copenhagen Economics and Professor of Economics, Aarhus University.
Despite the research suggesting NLP can be used to forecast FX, widespread cynicism remains - especially about attaining the ‘holy grail’ of an indicator which can successfully forecast an asset class as complex as a national currency pair, a cynicism which explains the former governor of the Bank of England, Sir Mervyn King’s oft-quoted remark, that trying to “forecast exchange rates,” was a “mug’s game.”
Similarly skeptical about AI’s contribution to forecasting rates is Marcus Roth, an Operations Manager at AI business intelligence provider Emerj:
“Business leaders in finance should not expect an AI to handle their foreign exchange trading anytime soon, and they might do well to be wary of AI vendor companies claiming to offer forex solutions.” Says Roth.
Indeed many text-based analyses of currency pair sentiment often fall short of providing an edge - possibly because much of the news about Forex is backward-looking. Such is the view of Copenhagen’s Lunde, for example, who’s study found no ex ante value in analysing sentiment for Yuan FX pair news - though studies that are more successful usually adopt an oblique approach focusing on sentiment for currency drivers rather than directly related news.
Yet the allure of trying to forecast foreign exchange remains powerful: in an article which appeared in the Financial Times called “
Top 10 Hits and Misses of Lord King’s Predictions
”, it was noted that even Sir Mervyn King himself finally succumbed to the temptation of forecasting Forex rates, when he said Sterling would “stay strong” - three years after his “mug” comment.
And, according to Emerj’s Roth, if a Forex trading solution using AI is to come about, it will probably emerge from ‘Silicon Valley’ and although RavenPack is technically headquartered in Marbella, Spain, it is part of a new breed of fintech company with close similarities to its Californian cousins.
Despite the skeptics and the persistence of belief in the random walk theory - which states prices are impossible to predict because they undertake a ‘drunken sailor’s walk’ across traders screens - there seems to be a lot of high-level research that the complete opposite is true, and Forex is forecastable, such as that put forward in an article published on the World Economic Forum’s (WEF) website, which tests the accuracy of the
IMF’s exchange rate forecasting model
“While movements of nominal exchange rates in the short term are notoriously difficult to predict, exchange rate models based on macroeconomic fundamentals can explain real effective exchange rate movements in the medium term surprisingly well,” says Pınar Yeşin, a Senior Economist at Swiss National Bank.
This suggests that by combining the short-term strengths of AI-generated news sentiment-based models with longer-term valuation models, such as the IMF’s, it might be possible to develop robust multi-timeframe predictive frameworks for FX.
Indeed, the research does seem overwhelmingly weighted towards suggesting there are ways to gain an edge on the market as only a simple googling of ‘forecasting exchange rates using AI’ or ‘NLP’ will attest.
One need only scroll through the list of search results to witness the volume of research claiming to be able to forecast Forex, to conclude, there probably is a good chance computers can give - even a mug like this - the upper hand.
You can easily implement the above research based on RavenPack data using the RavenPack Analytics Platform, which includes sentiment data on over 250,000 individual entities and events, available on visual dashboards or via web APIs. Request a trial today.
Please use your business email. If you don't have one, please email us at email@example.com.
We will process your personal data with the purpose of managing your personal account on
RavenPack and offering our services. You can exercise your rights of access, rectification,
erasure, restriction of processing, data portability and objection by emailing us at firstname.lastname@example.org. For more information, you can
Your request has been recorded and a team member will be in touch soon.
High inflation has returned in developed markets after decades of lying low. In our latest paper, we show how to build an inflation-based asset allocation strategy using sentiment data and we illustrate that sentiment-based strategies outperform models that depend merely on past observed inflation values.
This year's RavenPack Research Symposium brought two intense days of knowledge sharing in London and New York, from 25 top experts in natural language processing, quantitative investing and machine learning. Together, we explored how firms can leverage new language models to generate alpha, better manage risk and respond to calls for more sustainable investment practices.
Human capital is at the heart of value creation. Our latest research demonstrates how unprecedented workforce insights, sourced from over 200 million job postings, can generate more alpha.