Sentimental Journey

Sentiment is defined by Merriam-Webster as an attitude, thought or judgment prompted by feeling. Feelings are rarely mentioned in conjunction with technology, but much has been said lately about algorithmic trading using market sentiment – feelings – from social media sites such as Twitter.

The theory is that an algorithm mines online data and performs sentiment analysis using simple terms to try to qualify and quantify the emotional chatter around a particular market. It then gauges whether the feelings for a particular stock or commodity are negative or positive, and uses the information for making trading decisions.

Uniting technology and emotion is a fascinating concept, one that sometimes keeps me awake at night considering the possibilities. Markets are run by humans, and humans are emotional creatures. Humans crave acceptance, the need to be liked or be part of the in-crowd, and that can lead to sheep-like behaviour when it comes to markets. Paired with lightning speed trading technology, that behaviour can translate into extreme volatility.

On the day of the Flash Crash last year a large sell order placed in E-mini S&P 500 futures contracts on the CME sparked a totally human panic when fear was in the air and sentiment was already leaning toward the bearish. The panic was exacerbated by algorithmic trading strategies and HFT, which caused an unprecedented drop within minutes and wiped out $1 trillion in market value before recovering.

Other emotions come into play in markets; jealousy (or envy) that another firm is making money on a particular asset class can cause a bandwagon-like jump into unfamiliar territories or instruments. The resultant shopping spree can lead to market bubbles as prices rise like helium balloons – until reality finally pricks. This is when fear takes over, creating the desire to sell and get out while the going is good.

The question is, can these emotionally-led boom/bust cycles be predicted using Twitter-mining, market sentiment algorithms? As I said in an interview with Advanced Trading in April, I think you could use a Twitter algo to get a sentiment reading on a particular topic, whether it be revolutions or how people feel about the economy. The problem is that by the time you’ve got that information, it’s more of a trailing indicator rather than a leading indicator.

U.K. hedge fund Derwent Capital’s wager that it can predict market direction three or four days in advance with nearly 88 per cent accuracy using a Twitter algo is an interesting one. The algorithm could be successful if it were used to track consumer confidence or to gauge consumer reaction to new products and predict sales. But I think a sentiment algo is unlikely to be able to deal with unforeseen events, and these are increasingly common. Earthquakes, tsunamis, flash crashes, credit crises, volcanoes – no one did a very good job of predicting any of these nor did they accurately foresee the global fallout and repercussions.

Who would have predicted that the largest one-week drop in crude oil prices in history would occur just after Osama Bin Laden’s death? The markets saw his death as bearish since a source of geopolitical tension had been removed from the equation. The fact that Libya, an oil-producing nation, was still wracked by civil war – a much more bullish indicator – was of lesser emotional importance at that time.

Most sentiment-based algorithms employ various kinds of correlations. In the case of oil, for example, one could have been ‘if on Twitter the word explosion occurs X times in conjunction with Libya then buy crude oil’. In the days immediately following Bin Laden’s death this would not have worked very well.

Correlations are everywhere, but they are not always very good market indicators. The Anne Hathaway factor is a good example. The Huffington Post pointed out in March that whenever Anne Hathaway is in the news the stock price for Berkshire Hathaway goes up. (When Bride Wars opened, the stock rose 2.61 per cent.) Atlantic editor Alexis Madrigal found the theory enticing and asked me if I thought it was possible. I told him that we come across all sorts of strange things in our line of business, and one of them is strange correlations.

Madrigal said in the article: “The interesting thing, though, is that it’s all statistics, removed from the real world. It’s not as if a hedge fund’s computers would spit [out] the trading strategy as a sentence: ‘When Hathaway news increases, buy Berkshire Hathaway.'”

This is an amusing anecdote, but it does show how correlations using market sentiment can go wrong. Market-predicting formulas correlate everything against everything, and Anne might fit in there somewhere. Generally, though, the correlations are between some statistical indicator and a stock or industry.

Whether your information comes from a Reuters news story, exchange data, Twitter feeds or even Facebook it can be used in an algorithm. Your algo can include correlations between a famous movie star and a stock price, or weather predictions and the likelihood of a bad hurricane season. It can be as fanciful or as factually-based as you like. Whatever your choice, your algorithm will need constant adjustments to ensure it is not running in the wrong direction.  An algorithm can provide an interesting take on alpha generation, like Derwent Capital’s Twitter algo. Or it can contribute to an almighty flash crash if it is not under control.

Algorithmic trading is like bungee jumping – fun and scary with the potential for a painful crash. Keeping extra safety lines on your algos, the ability to change them on-the-fly and to monitor them for possible rogue tendencies, will ensure that you avoid the crash. A Twitter-based sentimental journey shouldn’t cost you the earth.