Whether you are a quant investor, a CFA, or you studied as an economics/business major in university, you will have heard of terms like ‘regressions’, ‘backtests’ and ‘r-squared’.
These are traditional statistical techniques used to develop predictive models to select stocks that are supposed to outperform the market – think smart beta products. They are also deployed to predict whether an index or an asset class will gain or lose the next day, week or month – think tactical asset allocation solutions. They work best if certain assumptions like normality or linearity in the data are met. And most of the time you can use Microsoft Excel to model these techniques.
But what if normal assumptions aren’t met? For example, one assumption that can cause issues with traditional techniques is when multiple variables – think factors – in a model exhibit multicollinearity, which means they are correlated among each other.
Machine learning covers a range of techniques that can help because it is more effective at dealing with complex situations.
The great leap forward
However, once you embrace machine learning, you also need to graduate from Excel and enrol into programming languages like R, Python and Julia, to name but a few.
The same quant or CFA working in the financial services sector, deals with traditional data sets such as stock prices, company balance sheets and income statements, and macroeconomic data like inflation and interest rates.
However, they may have heard of newer terms including Big Data which uses satellite images, tweets, sentiment in news or earnings conference call recordings and transcripts.
Machine learning, Big Data and AI are the buzzwords of the moment and are expected to reshape the future of investing.
The big question is whether the industry is actually applying these techniques and coming up with effective fintech solutions or just talking about it.
The CFA Institute recently surveyed a sample of charterholders which covered 700 respondents across the globe with a wide variety of titles from analysts to CIOs and portfolio managers. The results are loud and clear: very few are using the newer techniques, software and data.
In fact, only 10% of the portfolio managers who responded said they use AI or machine learning but 50% of them run backtests and use regression analysis. The results are similar for analysts: 75% of the respondents say that they don’t use AI, machine learning or Big Data for their analysis.
Obstacles on the road
What are the causes of this slow adoption by the industry? The CFA Institute report goes on to identify five potential reasons:
- Costs: especially those to buy clean and ready-to-use Big Data sets;
- Talent: AI-trained professionals – with degrees in mathematics, engineering, physics and statistics – are not a commodity and prefer to work at companies like Google, Microsoft and Alibaba where research is more advanced and widespread;
- Technology: this is a fast-moving space and keeping pace isn’t easy especially if numbers 1 and 2 are scarce;
- Vision: strategic vision and leadership are a necessary condition;
- Time: implementation of a Big Data and AI infrastructure doesn’t happen overnight. Time and patience are key to success.
Until ‘fin’ (quants, discretionary managers, analysts ) meets ‘tech’ (chief technology officers, engineers etc), it will be difficult to make any progress.
But some pioneers are emerging. The CFA Institute report goes on to show some examples of solutions and strategies used by companies that are squaring up to the challenge:
- Trading execution, an area of market microstructure research. Machine learning algorithms are used to pick the cheapest and quickest route to market.
- Sell side research where Big Data sets can be used to enhance it. For example, geospatial information provides better insights into how companies operate and grow, as well as their competition, which is important for earning estimates.
- Stock picking with the aid of natural language processing. This can be applied to earnings conference call transcripts to assess management language around: omissions and obfuscations – are they trying to hide something?; spin – is the content exaggerated and overly scripted?; and blame – are they diverting responsibilities?
This article originally appeared in the February edition of Citywire Selector magazine.
Credit: Google News