Tuesday, June 5, 2012

It Pays to Watch Insiders

We recently performed a complex event analysis with our data. In this case we were looking at insider purchases that follow layoff events and the effect of these joint events on market performance. Purchase events on their own typically forecast a slight rise in a stock. Layoff events can be difficult to interpret. Are they the sign of fundamental problems or a sign of correcting fundamental problems? However the joint signal of layoff plus purchase does appear to have predictive power. In short, we performed an event study around the 33064 insider purchases in top 1000 US companies between 2007 and 2011. 4737 of these events occurred within 2 weeks of a layoff event. We found that the market adjusted returns of these companies outperformed the S&P by 90 basis points and the other insider purchases by 55 basis points in the following 40 trading days.

We built and back tested a simple trading strategy based on these events as well. Start with $1,000,000. Every time a joint layoff/purchase event occurs we invest $10,000 in the stock. We hold for 20 days and assume 40 basis points round trip trading costs. The performance is below and it outperforms the S&P by 30% over this time period.

Wednesday, March 28, 2012

Anticipating Volatility with Future Events

We’ve spent a lot of time in this space exploring relationships between media sentiment and direction of future stock moves. In this post, we’ll take a look at a relationship between anticipated future events and future volatility.

What that chart shows is that we see a strong and statistically significant increase in market-relative realized stock volatility around anticipatable future events, even when controlling for whether or not these events are earnings related.

Using Recorded Future’s advanced temporal language processing technology, we have extracted all forward looking statements relevant to S&P 500 companies from thousands of web sources over a four year period. That is, we look for phrases like “Apple is expected to release its new iPad next Wednesday,” or “Settlement expected in the Google/Oracle patent case on April 10th,” and associate them with the anticipated future event date, and the company(s) expected to have news on that day.

By aggregating and filtering to these future events that we think are likely to happen, we end up with a total of 7548 company/day pairs that are not on earnings days, and we see about a 10% bump in realized volatility on the event day relative to the previous 3 weeks. After the event has taken place, we see a statistically significant reduction in volatility versus the period before event time.

If you’re interested to hear more about this result, or are interested in incorporating knowledge of future events into your investment process or product, contact us.

Friday, March 9, 2012

Media Factors as Overlay on a Mean Reversion Strategy

In the last post we introduced a new dataset and revisited a stand alone trading strategy based on news attention. We've gone a step further and considered using news attention as an alpha overlay to an existing strategy. For the baseline strategy, we've started with a contrarian mean reversion strategy presented in Lo and MacKinlay (1990). We applied this as a daily strategy where we buy the stocks that underperform the market today and short the stocks that outperform the market. We rebalance at the closing price daily. We weight the stocks in the portfolio according to the difference between the stocks daily return and the market return with the biggest under and overperformers receiving the largest weights. In backtesting, this baseline strategy returns 167% percent since 2007 with a Sharpe Ratio of 1.09 (pre trading costs).

Our theory was that stocks with high levels of attention were less likely to revert than other stocks. So we modified the baseline strategy to remove the stocks in the top and bottom deciles of our media analytic factor (stocks with most positive attention and most negative attention) from the daily portfolios. This strategy yielded a 288% return with a Sharpe of 1.59. The annualized alpha due to adding the RF factor to the basic strategy is 10%. This is an illustration of how you could use a media analytic factor to improve the performance of a sample statistical arbitrage approach. These approaches are typically unaware of the news about the companies in question and adding in news factors in an appropriate way can increase performance.

The annualized alpha due to adding the RF factor to the basic strategy is 10%. This is an illustration of how you could use a media analytic factor to improve the performance of a sample statistical arbitrage approach. These approaches are typically unaware of the news about the companies in question and adding in news factors in an appropriate way can increase performance.

Wednesday, March 7, 2012

Revisiting Media Based Trading with a Broader Data Set

We've recently started rolling out a new dataset with a tenfold increase in the number of sources and an additional two years of history. We are pretty excited about the additional breadth of coverage and now having a historical archive of five years. The first thing we tried with this new data was a version of the deciling approaches we've shown earlier on this blog. Having tens of thousands of sources allowed us to do some source filtering and we focused on news based sites with high web ranking. Using a similar approach as the earlier analysis, we show the returns of the portfolios constructed from the different deciles of sentiment weighted attention. The companies in Decile 10 have the most positive news for the day and the companies in Decile 1 have the most negative news. The returns are based on one day holds.

Again following the earlier analysis, we constructed a long/short portfolio buying the stocks in the tenth decile and shorting the stocks in the first decile. The backtesting performance of this portfolio was quite good, having an overall Sharpe Ratio of 4.

When we compare these strategies to conventional Quant trading strategies (Fama-French Value and Momentum portfolios) we see very low correlations near 0.05.

We will soon have additional analyses that use this new broader data.

Monday, February 27, 2012

Recorded Future - Enabling Human Traders with Quantitative Signals

Powered by Recorded Future. That's the news from Max Bowie, editor of Inside Market Data.

We recently went public with a new partner -- Titan Trading. They'll incorporate Recorded Future sentiment, media analytics and event data in their Tick Analyst platform: "so we can serve up sentiment changes and upcoming actions within our signal stream." 

Titan's taking the lead with developing signals from Recorded Future data for human traders. We're excited for their launch and even more so for what they've done with the Recorded Future dataset.

Tuesday, August 9, 2011

Out of Sample Strategy Performance Update

A little while ago, we published a blog post on a trading signal we've developed internally based on media analytic data. In May, we launched a live version of the components of that signal, as a feature of the Recorded Future API. Our customers can pull this data directly from the API at 3:30pm, giving them time to trade before the equity markets close at 4.

Taking the same strategy we presented earlier, and using the live data as it was available to our customers at 3:30, we have rolled our backtest forward, and looked at the performance of this strategy over the last few tumultuous months. Between May 13, and August 5, this strategy returned 10.4%, while the market lost 9.9% of its value. These returns have been fairly consistent, and turnover has been similar to what we saw in our original backtests. The results are plotted above.

Of course, this is a short time window - encompassing just 59 trading sessions, and we haven't taken into account trading costs in this analysis. Still, we find these results encouraging and will continue to look for other sources of long-term signal in our studies going forward.

If you'd like to learn more about the Recorded Future media analytics API, contact our team.

Friday, March 4, 2011

Factor Modeling Media Analytic Data

At Recorded Future, we’re scouring the web for predictive signals in online content. Previously, we’ve covered our efforts at complex event modeling, and liquidity modeling using news flow information. Publicly, we’ve also touched briefly on some of our returns modeling - we’ve seen instances of particular blogs that seem to have superior predictive power in terms of their ability to write about stocks that will outperform.

Recently, we’ve expanded this approach to build a whole-market factor model that uses media analytic data to predict excess returns. Using aggregate data for the S&P 500, which is available to our API customers, we’ve built a number of factors that are derived from online sentiment and momentum of S&P 500 constituents that show statistically robust predictive signals of market-relative returns over a 1-day to 1-week investment horizon in a time-series cross-sectional modeling environment.

Factor Examination
Let’s take a look at one such factor, which is based on sentiment and momentum. If we take this factor, and break it into deciles by day and then construct portfolios for each decile, we see the following cumulative continuous returns in these portfolios. We’ve included dividend-adjusted returns to the SPDR S&P 500 ETF (SPY) as a benchmark in bright orange.

You can see quite clearly that over the last two years, our top decile (in orange) has outperformed all other deciles in a fairly consistent manner. Meanwhile, the bottom three deciles (the three darkest shades of blue) have underperformed all other deciles, as well as the market. One thing to note is that this relationship is not strictly linear. For instance, our 2nd, 3rd, and 4th place deciles actually fall near the middle of the returns distribution, which may have something to do with the construction of this particular factor.

If we compare the portfolios to the performance of the S&P 500 over this period, we find that the portfolio in the top decile has a Beta of 1.08, assuming a risk free rate of return roughly equivalent to that of T-bills over the period. It has a statistically significant annualized (continuous) Jensen’s alpha of +16% over the period. When we examine the bottom two deciles under the same assumption, we see that they are high Beta portfolios (1.37 and 1.34, respectively), but with statistically significant and negative alphas, at -42% annually, and -26%, annually. As you might imagine, constructing hedged portfolios out of the securities in these deciles provides some possibly compelling trading strategies.

If you’d like to experiment with this approach yourself. We’ve made some R code available on our Google Code site which will pull in market data, Recorded Future data, and perform this sort of decile analysis on a factor of your choosing. You’ll need a Recorded Future API token to pull that data.

Soon, we’ll discuss the inclusion of a factor like this into a portfolio built using other factors based on Recorded Future media analytic data, and find out whether a portfolio like this can stand up to trading costs, and evaluate its performance in an out-of-sample context.