FX Trading Signals From Systematic News Flow Analysis

By Peter Hafez and Junqiang Xie of RavenPack Quantitative Research

 

In this study we test a short-term FX trading strategy that uses the principles of technical analysis to create buy or sell signals based on data derived from fundamental news. Short and long term sentiment inflection points are captured by consulting a set of sentiment indexes that measure the trailing sentiment on both scheduled and unscheduled economic and geopolitical news events.

The trading signals generated systematically from using sentiment indexes can give an effective directional forecast for the EURUSD valid for some hours after an inflection point.

 

The RavenPack U.S. Macro Sentiment Index Predicts EURUSD Price Movements following Sentiment Inflection Points

  • With a 10 hour holding period, a cross-over strategy between short (1-week) and long-term (3-month) sentiment generates an annualized Information Ratio of 1.61 based on monthly P&L.
  • With a 3 hour holding period, a cross-over strategy between medium (1-month) and long-term (3-month) sentiment generates an annualized Information Ratio of 1.29 based on monthly P&L.

 

RavenPack Data

The RavenPack U.S. Sentiment Index is based on RavenPack’s News Analytics 3.0 dataset which systematically tracks and analyzes information on over 138,000 key geographical entities, more than 2,200 government organizations, 150 major currencies, 80 traded commodities and over 30,000 companies. In addition, RavenPack covers over 1,200 events of which 895 relate to unscheduled news such as political events, natural disasters, wars, etc., as well as scheduled news such as the release of important macroeconomic indicators. For any detected news event, RavenPack generates an Event Sentiment Score (ESS) signaling its potential impact on any given economy or financial market.

 

1. Introduction

Typically, technical analysis is applied to continuous data sets like price and traded volume, while fundamental analysis is based on discrete observations of the state of an economy which are updated at a much lower frequency. Purist technical or fundamental analysts will not mix the two techniques, but many market participants use the two in combination – often with fundamentals determining direction and conviction and technicals confirming the timing of entry and exit.

 

In this paper, we present a simple cross-over strategy that tracks the short and long term[1] sentiment trends for a given economy in order to detect cross-over events identified as sentiment inflection points. We take a “bullish” view on a currency when the short term sentiment trend crosses above the long term sentiment trend and a bearish view when the short term sentiment trend crosses below the long term sentiment trend.

 

To measure the sentiment trend for a given economy (the United States, for example), we take the moving average of all news events with ESS scores over a certain aggregation window. More specifically, we examine three different aggregation periods including 1-week, 1-month, and 3-month windows. All economy-specific macro sentiment indexes are updated for every minute when non-neutral news events are recorded. To detect sentiment inflection points, we find the cross-over between the short-term and long term indexes. Specifically, we determine the following three cross-overs: 1-week vs. 1-month; 1-week vs. 3-month; and 1-month vs. 3-month. While this strategy can be applied to multiple currencies, here we only focus on U.S. events in a cross-over strategy on the EURUSD exchange rate. More specifically, we take a short position in the EURUSD when the short-term U.S. macro sentiment index crosses above the long-term sentiment index, suggesting a bullish view on the U.S. Dollar. Conversely, we take a long position in the EURUSD when the short-term index crosses below the long-term index.

 

2. Evaluating the Cross-Over Strategy

To test the cross-over strategies, we use 1-minute bar prices[2] over the period of January 2010 to June 2012, and evaluate signal performance up to 10 hours after the cross-over event. The strategies are evaluated based on the annualized Information Ratio of the monthly P&L assuming fixed daily capital. Fig 1 below plots the Information Ratio by holding horizon (1-10 hours) for each strategy based on non-overlapping signals[3]. We find that the cross-over strategy generates strong performance during the back-testing period. The highest annualized Information Ratio achieved is 1.61 and is observed for the 1-week vs. 3-month cross-over strategy with a holding period of 10 hours. The 1-month vs. 3-month cross-over strategy generates an Information Ratio of 1.29 based on a 3-hour holding period.

Fig 1:  Annualized Information Ratio of Monthly P&L by Holding Period; Jan 2010-Jun 2012; Non-Overlapping

This figure shows the annualized Information Ratio of the monthly P&L assuming a fixed daily capital of $10,000 by holding horizon (1-10 hours) based on all signals from January 2010 to June 2012.

SOURCE: RavenPack, Histdata, January 2013

 

Over the back-testing period, the 1-week vs. 1-month strategy generates about 5.64 trades on average per month, the 1-week vs. 3-month strategy generates about 5.94 trades per month, while the 1-month vs. 3-month strategy generates about 3.15 trades per month[4].

 

Fig 2 lists the Hit Ratios (the proportion of winning trades) of the non-overlapping signals for each strategy by holding horizon (1-10 hours). Overall, the highest Hit Ratio is 0.59 and is observed for the 1-month vs. 3-month cross-over strategy with an 8-hour holding period. For the 1-week vs. 1-month strategy, the highest Hit Ratio is 0.51 when the holding period is 10 hours. For the 1-week vs. 3-month strategy, the highest Hit Ratio is 0.55 when the holding period is 10 hours. Consistent with the pattern of the Information Ratio, the 1-week vs. 1-month strategy tends to have a lower Hit Ratio across all holding periods. Over the back-testing period from January 2010 to June 2012, the 1-week vs. 1-month strategy generates about 5.64 trades on average per month, the 1-week vs. 3-month strategy generates about 5.94 trades per month, while the 1-month vs. 3-month strategy generates about 3.15 trades per month6.

Fig2: Hit Ratio of Non-overlapping Trading Signal

This figure shows the Hit Ratio of the non-overlapping signal for each strategy by holding horizon (1-10 hours). The average number of trade per month for each strategy from January 2010 to June 2012 is shown at the bottom of the table.

SOURCE: RavenPack, Histdata, January 2013

 

Fig 3 plots the cumulative pips generated from the 1-week vs. 3-month cross-over strategy for a holding period of 10 hours. As can be observed, the 1-week vs. 3-month strategy is able to generate stable performance over the backtesting period.

 

Considering the average number of pips earned per trade, not surprisingly we find that as the holding horizon increases, the pips earned per trade tend to go up. For the 1-week vs. 3-month strategy, the maximum number of pips earned is obtained after 10 hours with 9.0 pips, while the 1-month vs. 3-month strategy delivers 7.5 pips after 9 hours. Finally for the 1-week vs. 1-month strategy, the maximum number of pips earned is obtained after 10 hours with 3.6 pips

Fig 3:  Cumulative Pips For the 1-Week VS 3-Month Strategy (Non-overlapping)

This figure plots the cumulative pips earned from the non-overlapping trading signals of the 1-week vs. 3-month cross-over strategies with 1 10 hour holding horizon, from January 2010 to June 2012.

SOURCE: RavenPack, Histdata, January 2013

 

Overall, we find that the news based cross-over strategy is able to predict future exchange rate movements over short-term trading horizons with positive annualized Information Ratios and stable returns over the back-testing period.

 

3. Conclusion

In this paper, we introduce a hybrid Forex trading strategy that applies simple technical analysis to fundamental news. Specifically, we consider a cross-over strategy that tracks the short and long term sentiment trends for a given economy in order to detect cross-over events that can be considered likely sentiment inflection points. The economy-specific sentiment indexes are constructed based on economic and geopolitical news events available as part of the RavenPack News Analytics dataset. We take a “bullish” view on a currency when the short term sentiment trend crosses above the long term sentiment trend and a bearish view when the short term sentiment trend crosses below the long term sentiment trend.

 

Applying this strategy using one minute bar data of the EURUSD exchange rate from January 2010 to June 2012, we obtain strong performance results with holding periods of up to 10 hours. With a 10-hour holding period, a cross-over strategy between short (1 week) and long-term (3 months) sentiment generates an annualized Information Ratio of 1.61. With a 3-hour holding period, a cross-over strategy between medium (1-month) and long-term (3-month) sentiment generates an annualized Information Ratio of 1.29. These findings suggest that RavenPack’s news analytics of both scheduled and unscheduled global macro news can be used as a new source of alpha for market participants in the FX market.

 

 

 

 

 

A more detailed white paper on this strategy and approach can be downloaded from http://ravenpack.com/research/Intraday-FX-Trading-Using-Sentiment.pdf

[1]Faster vs. slower moving sentiment average.

[2]1-minute bar data on the EURUSD exchange rate is downloaded from http://www.histdata.com

[3]  We ignore any new signal that appears during the holding period.

[4] For the non-overlapping 1-week vs. 3-month strategy 83.4% of days have no trading, while the equivalent numbers for weekly and monthly are 31.5% and 9.4%, respectively. For the non-overlapping 1-month vs. 3-month strategy, 90.3% of days have no trading, while the equivalent numbers for weekly and monthly are 68.5% and 25% respectively. For the non-overlapping 1-week vs. 1-month strategy, 84.8% of days have no trading, while the equivalent numbers for weekly and monthly are 35.4% and 9.4% respectively.



A related article on this website can be found at http://www.hedgefundinsight.org/an-example-of-a-news-based-trading-strategy/