lternative data. It was supposed to be the modern day alchemy of Wall Street – turning boring data into golden alpha. There has been an explosion in alternative data over the last few years from all corners of the industry, as individuals sought to turn data into gold, but there are interesting (and I believe unforeseen) issues emerging. Read on.
Everyone has been involved in the rush. Brokerages have been looking at their own data sets to see what they can provide their clients that may have been historically overlooked, with firms offering up retail banking transaction data, loan portfolio data, and anything else they think might be valuable. Startups have entered the vacuum to harness and sell exotic and remote data sets such as private jet leases and satellite imagery. The incumbents, such as Bloomberg and NASDAQ, have launched data marketplaces. Funds have given it all a try, while attempting to find their own alternative data sets, such as Twitter stream analysis and speech parsing.
As always, everyone is looking for alpha, and alternative data seems to be a potential holy grail. I was in the office of a major fund just over a year ago (how nostalgic), when the individual I was meeting with wanted to show me a new tool they had created. The application ‘listened’ to investor calls, and would try to interpret the ‘mood’ of the speaker based on words and tone – did the speaker sound worried? Confident? Excited? The application would try to summarize pieces of the narration, the tone, and the word usage with sentiments that described the speaker’s mood, then pass that information on to an aglo that could trade based on the outputs. I know another firm that tried to predict the US unemployment rate by consuming the entire Twitter firehose, then analyzing how many people had referenced a new job vs. how many people had referenced being fired. Everyone is looking for an edge in the data.
But what happens once you start analyzing all these new data sets? Some of them may change. There are two effects that came to mind that I believe are active in the market.
First, there is an effect in the world of physics called the Observer Effect. It speaks to the fact that systems can be disturbed by the simple act of observation. When you look at something, measure it, analyze it, or otherwise interact with it (even look at it), you can change it. Think about measuring the air pressure in a tire – letting a little air out to measure it actually changes it. What data sets might be changing just because we are starting to measure them?
Second, for all those who remember their high school physics, there is Newton’s Third Law – that for every action in nature, there is an equal and opposite reaction. What will the reactions be to using these data sets in investment decisions?
We’re now seeing some answers to those questions in the market.
Bloomberg recently ran a piece that CEO’s are now adapting the words they use to cater to text parsing systems. As management knows more and more institutional clients are using alternative data tools such as transcript analysis, they are changing how they communicate to the world – but what did we expect? If I were the CEO/CFO/COO of a public company, I would run my remarks through some of these tools before the call, just to see which fared best, and gave the best signals to others using those same tools. If investors are using systems to analyze my update in order to decide whether to buy or sell shares in my company, why not just see what the outcome is before I even have the call?
The challenge only grows. If a significant portion of the market is using these signals, then the impacts can become distorted. What happens when all management teams start monitoring their verbiage or tone? The signal loses its impact, or even worse, creates unintended consequences – like training algos on datasets that are now incorrect. Regardless, we may get to the point where investor calls are even more preconstructed than ever before, with only certain words, said in certain ways, allowed.
This is likely just the beginning. We’re now seeing the impact of these two effects on alternative data: what changes just by observing it, and what changes as a reaction to the original force. As brokerages and funds look for alternative data to drive their investment decisions, we need to stop and ask ourselves what the knock on effects might be. How might using that data set actually change the underlying data set, or how might the dataset react to being used/monitored. These are important questions to track as we continue on into the alternative data universe.
Nothing in the world is static – and everything changes – but how exactly is it changing? That’s a question for the alternative data ecosystem to watch closely.
Best of luck out there,