RavenPack Sentiment and Macro-Economic Indicators

The relationship between market sentiment and macro-economic indicators has always been an intriguing topic amongst newspaper columnists, academics, economists, and policymakers; especially when the economy is going through a recession. Calculating sentiment from financial news is a more timely method than traditional macro-economic indicators and thus could potentially improve trading models that build on macro-economic variables. Considering the correlation between the quarterly change of the macro-economic indicators and the quarterly change of RavenPack’s Sentiment Index, we arrive at the following key results:

1. The RavenPack Sentiment Index is consistently positively correlated with the macro-economic indicators applying zero lag (including Real GDP, Inventory to Sales Ratio, Inventories, Retail Sales, the Michigan Consumer Confidence Index, and Personal Consumption Expenditures).
   
   
2. The contemporaneous correlation between the quarterly change of the RavenPack Sentiment Index and the quarterly change of real GDP is 56%. Applying up to three lags improves the overall correlation to nearly 70%, with a 87% correlation since 2005.
   
   
3. The RavenPack Sentiment Index is highly correlated with the University of Michigan Consumer Confidence Index (41%) - suggesting the complementarities of the RavenPack Sentiment Index to this traditional sentiment measure.
   
   
4. The contemporaneous correlations with the macro-economic indicators are much higher for the RavenPack Sentiment Index than for the University of Michigan Consumer Confidence Index - suggesting that the RavenPack Sentiment Index could be a better sentiment measure.
   
   
5. The lagged RavenPack Sentiment Index quarterly change is highly correlated with real GDP and inventory changes - suggesting a potential forecasting power of the RavenPack Sentiment Index for the macro-economy.

In Fig 1, we have plotted the real vs. predicted U.S. GDP growth since January 2001 through July 2011.

News Sentiment: Highly Correlated with the S&P500

As part of our ongoing research effort in this area, we continuously try to learn more about how to better capture sentiment trends at company, sector or market level. In this latest study, we have come up with a simple, intuitive, and yet robust approach to capture the sentiment trend on the US market. Below, we have shown the RavenPack Sentiment Index mapped against the cumulative return of the S&P500.



Here are some of the key results from the study:

The RavenPack Sentiment Index Moves Closely with Financial Markets

• From January 2000 to September 2011, the contemporaneous correlation between the RavenPack Sentiment Index and the S&P500 Index is 79%

• The RavenPack Sentiment index is consistently highly correlated with the S&P500 Index across different market trends. Especially, we find an average correlation of almost 90% during bear markets

The RavenPack Sentiment Index is both Statistically and Economically Significant

• A statistically significant causal relationship exists from market sentiment to stock market returns

• The sentiment trading strategy based on monthly VAR(2) yields an annualized return of 10.2% between March 2000 to September 2011

• The recursive monthly VAR(2) model is able to generate an out-of-sample annualized return of 6.7% between April 2006 and September 2011

• The sentiment based trading strategy based on weekly VAR(10) yields an annualized return of 13.4% between March 2000 to September 2011

• The recursive weekly VAR(10) model is able to generate an out-of-sample annualized return of 17.5% with an Information Ratio of 0.81
  

Research Webinar: Trading Strategies Using News Analytics and Stock Lending Volume

I thought you might enjoy learning about our next research webinar that will discuss trading strategies using news analytics and stock lending volume.

At the webinar, Dr. John Kittrell from Knightsbridge Asset Management will discuss key findings from his latest study entitled "Behavioral Trends and Market Neutrality" including:

1. How to use news sentiment signals to outperform the equity market
2. Enhance risk adjusted returns combining news analytics with diversifying stock lending metrics
3. Construct low turnover portfolios using signals from these sources

The event will take place on Wednesday, November 9th at 10:30am - 11:30am EST.

To register, please visit: https://www1.gotomeeting.com/register/989422168

Watch Out For Those Unexpected Events - Webinar Replay Video

I've received a few requests for a replay video of the research webinar. Below is a link to the "Watch Out For Those Unexpected Events" held earlier this week:

Research Webinar on Unexpected Events

I've been asked to present at RavenPack's upcoming research webinar on unscheduled news events. I'll be discussing some market response techniques and event-driven trading strategies. Other presentations will focus on the impact of macroeconomic and geopolitical events in trading. Here's a link to the research agenda and registration page: https://www3.gotomeeting.com/register/737465358

Event Timing and High Frequency Trading

The timing of company specific news events may impact short-term price discovery. This relationship becomes important for a high frequency or short-term trader to understand when events typically occur (e.g. pre-market, during market-hours, or while they are fast asleep).

To learn more about the timing of different events, I consider a set of event categories as detected by RavenPack. In total, there are about 330 event types all mapping into 21 broader groups. For most of these groups, I find high activity pre-market between 6:00 - 9:30 AM Eastern Time (EST), as well as during market-hours between 9:30 AM and 4:00 PM. Furthermore, many events occur after-hours between 4:00-8:00 PM.

Looking at Figure 3, primarily three groups stand out including insider trading, regulatory related events, and credit ratings that tend to occur mostly during market-hours. Looking closer at the insider trading group, I find that very few events occur pre-market for both insider-buy and insider-sell. The main difference between the two categories is that insider-sells are announced during after-hours, while insider-buys are announced post-market between 8:00 PM and 0:00 AM.



Compared to unscheduled events such as reorganizations, mergers, or strikes; scheduled events tend to occur more often pre-market or after-hours. More specifically, I find that about 48% of scheduled events take place pre-market compared to 40% for unscheduled events. During market-hours, the numbers are 23% and 40% for scheduled versus unscheduled events, respectively. While 29% of events during after-hours are considered scheduled, the equivalent value for unscheduled events is only 17%.

Given these findings, I think the price discovery process is different for scheduled than unscheduled events, not only because the latter has more of a surprise element to it, but also due to when in time these events typically occur.

More on this topic is available in my latest research study: Event Trading Using Market Response.

Sentiment Analysis More Timely Than GDP Figures

Gross Domestic Product (GDP) is considered one of the most important economic indicators for "taking the pulse" of the economy. Initially, GDP figures are only available as estimates, which can take years to get finalized. The first estimate of GDP is available about 1 month after the end of each quarter. This is followed by significant successive revisions for up to a year after the initial figure.

To get an early indication of where the economy is going, I find probing financial news to be more effective than using lagging economic indicators like GDP. For starters, news information about publicly traded companies is available in large volumes and in real-time. There are no complicated surveys, bureaucracies, or lag times, the information is processed as soon as it becomes public. By focusing on highly relevant company events in the news and by aggregating its sentiment, I make a move from micro to macro level economics.

The figure below depicts the quarterly changes in GDP and the Market Sentiment Index for the U.S, which shows a positive relationship between them. To test for significance, I consider a linear regression with sentiment being the explanatory variable. Furthermore, I consider whether there might be a lagged relationship between the two.



The analysis confirms a strongly significant positive relationship between US news sentiment and GDP (at a 99.9% level). Furthermore, I find that the strongest relationship is found without applying a lag. An overall R-squared is found to be 0.25 resulting in a correlation of about 50%.

The sentiment index is constructed by probing news facts on more than 7,000 U.S. companies in real-time. It picks up news on layoffs, earnings, management changes, and hundreds of other key events. The data derives from the micro level while the index formulae produce a macro perspective. There are no macroeconomic indicators in the construction of the sentiment index. In contrast, the GDP figures above were available with a lag of at least 4 months. Real-time news sentiment shows promise of being a more timely economic indicator than traditional government GDP figures.