The founders of MarketPsy Capital have developed a systematic investment strategy that arbitrages sentiment-based stock mispricing by analysing the
The founders of MarketPsy Capital have developed a systematic investment strategy that arbitrages sentiment-based stock mispricing by analysing the language of stock-related news and information such as financial news, company press releases and SEC filings. Richard Peterson, co-manager of the MarketPsy Long-Short Fund, explains how the firm’s strategy evolved.
HW: What is the background to your company and fund?
RP: Building on prior academic research in behavioural finance and linguistics, MarketPsy Capital has developed a systematic investment strategy that arbitrages sentiment-based stock mispricing. Our research led us to develop proprietary software that collects, organises and analyses the language used in stock-related news and information text such as financial news, company press releases, SEC filings and other sources.
Such text is an excellent source for evaluating the sentiment of a company’s executives or its major investors. Using machine-learning algorithms, we have found the optimal weightings of such sentiment for predicting future stock returns and identified various robust trading signals that together determine the MarketPsy portfolio.
You can’t help it, but your brain (and everyone else’s) is wired to sabotage your investing. I discovered this through my own experiences in the mid-1990s. As the senior project in my electrical engineering coursework, I chose to develop quantitative neural network-based stock index forecasting software. I found the software to be somewhat predictive of the markets, so I decided to trade the software’s forecasts with a small managed account.
Something unexpected happened. Days when I was most reluctant to take a trade signal were the times that the most profitable trades were made. I measured this effect over three years, and then extended it to my read on ‘the mood of the market’. The pattern was consistent over the years – the market’s mood was inversely correlated with the future direction of prices. High media negativity was correlated with future price gains over the next week.
I realised that an enduring source of alpha lies between our ears. Understanding the workings of the brain, and in particular how investment decisions are made, unlocks a trove of novel investment strategies. And the clues to how and what investors think lies in the unconscious (their feelings) as well as in their conscious minds (what they say in conversation).
If there is alpha in understanding the mood of the market, how can we access the minds of thousands of investors to test this – and potentially profit? I experimented at Stanford University with functional magnetic resonance imaging technology, facial electromyography, EEG, and psycho-physiological tools for understanding investor decision-making, but these tools are impractical due to cost, discomfort and limited scalability.
In short, we couldn’t do this directly, so we had to look for evidence of investors’ mindsets through their behaviour and conversation. Stock message boards, where investors congregate and discuss stocks, seemed the first place to turn for large amounts of easily downloadable text that represented the thoughts of millions of investors. Investor communities, where investors trade and discuss simulated and real portfolios, are another location to access portfolio decisions, analysis and discussions.
At the same time as I was performing neuroimaging experiments and studying the psychology of investing, Jacob Sisk was a lead researcher at Codexa and Leinweber & Co., two financial text analysis start-ups founded by David Leinweber. Their research unearthed alpha in the text and numerical data in SEC filings and stock message boards.
Jacob co-authored an academic paper on this topic (Financial Communities) with Sanjiv Das at Santa Clara University. Jacob and David pursued the hedge fund start-up process in 2004, but turned away from launching a fund and began consulting for other firms, including BGI and Cooper Neff. Between 2004 and 2007 Jacob worked as a researcher at Yahoo!, where he identified the behavioural predictors of non-financial activities such as dating and advertising clicks.
After working with several software development teams in 2004 and early 2005, I met Yury Shatz and employed his firm for software development, mining text and creating stock forecasting systems. Yury had modelled fisheries collapses while working at the United Nations Food and Agriculture Organisation in Rome for six years, and using his statistical modelling background he took a personal interest in the task I was working on. Yury self-initiated expansion of the scope of the work we were doing, and we committed to creating a fund using the software he was developing.
In 2005 David Leinweber guided me in establishing a real-time portfolio that, with Yury’s programming expertise, began trading in June 2006. This trading system measured the total amount of ‘fear’ words in the stock message boards of 30 arbitrarily chosen large stocks. The two stocks with the highest level of ‘fear’ were held long, and the two with the lowest were held short.
Every day emails were generated with the results of the daily analysis. The average holding period was nine days per stock. As this strategy was demonstrating performance in its first year, David put me in touch with Jacob Sisk in mid-2007. We immediately recognized our shared passion and began planning for the launch of MarketPsy Capital with Yury, Jacob, and I as the managers.
As text became easily available online, several academic researchers joined the text-analysis research field. Papers by Antweiler and Frank revealed that sentiment detection in the stock message boards could predict implied volatility of stocks. Paul Tetlock et al found that the level of negativity in Dow Jones News stories is predictive of the future direction of those stocks. Li Feng et al discovered that the sentiment of SEC filings is predictive of future corporate performance.
With our own simple ‘fear’ portfolio generating 17 per cent annual returns over one and a half years, and support from several academic papers, we decided to launch a hedge fund to exploit psychological alpha in financial text. In February 2008 Yury, Jacob and I incorporated MarketPsy Capital and began the start-up process. The MarketPsy Long-Short Fund was launched on September 2 and delivered returns net of 2/20 fees of 24 per cent over the first seven months since launch.
HW: Who are your key service providers?
RP: Our prime broker and custodian is JP Morgan Clearing, execution is provided by Goldman Sachs Execution and Clearing via REDI, and our introducing broker is Shoreline Trading. Our fund administrator is PanOptic Fund Services, legal counsel is Investment Law Group and auditor is Spicer Jeffries.
HW: How and where do you distribute the funds? What is the split of your assets under management between institutional and private clients?
RP: We have a BVI-domiciled master fund and both US- and offshore-domiciled feeder funds. High net worth investors account for 80 per cent of the fund’s assets and institutions 20 per cent.
HW: What is your investment process?
RP: Every day we receive trading signals on approximately 20 mid- and large-cap US stocks, which are used to create a beta-balanced portfolio for the fund. In the nine months since we launched the market-neutral MarketPsy Long-Short Fund, this quantitative strategy has delivered a 51 per cent net return.
Hypotheses about the psychological drivers of markets are generated and backtested on our quantitative linguistic data. Strategies are implemented as trading algorithms if testing and real-time monitoring is successful. The fund is entirely systematic.
HW: What is your approach to managing risk?
RP: We maintain position size limits of no more than 3.5 per cent for stock over USD5bn, 2.5 per cent for stocks with market capitalisation between USD1 and USD5bn, and 1.5 per cent for stocks with market capitalisation of USD250m to USD1bn. Industry exposure is limited to 20 per cent. We are beta neutral and do not use more than two times leverage, with an average of between 1.5 and one.
HW: How has your recent performance compared with your expectations and track record?
RP: Our performance has been consistent with our expectations. We do not expect any change going forward?
HW: What opportunities are you looking at right now?
RP: We are looking at new sources of linguistic information.
HW: What differentiates you from other managers in your sector?
RP: We are significantly ahead of the competition. We trade in all US equities greater than USD250m market capitalisations. We are experts in behavioural finance, computer science, linguistics and search engine technology.
HW: Do you have any plans for other product launches in the near future?
RP: We are considering a directional fund.
