By CFA Society

Access to computational processing capabilities far exceeds what we could have dreamed of a few years ago. Easy access to servers (the cloud) gives us endless possibilities that would have seemed unthinkable not so long ago. While there are industries that have leveraged more on this, others still have many possibilities to explore.

Quantitative finance and its application in the construction of investment portfolios represents an area with tremendous potential. represent an area with tremendous potential.

I'm not just talking about large language models (LLM), but also financial and statistical models that allow to maintain an investment philosophy from a systematic approach. from a systematic approach.

A universe of opportunities
Quantitative investment strategies are as varied as conventional ones. Some funds trade intraday, others long term. Some focus on value attributes, others on trends, and others on investor psychology or market biases.

A clear example available to all is Joel Greenblatt's famous "Magic formula" by Joel Greenblatt^,1 which proposes a systematic analysis that helps to build a portfolio with a value approach (Value Investing). And if we want to look at the origins of this philosophy, Ben Graham speaks of Net-Nets ^{2 3} as companies in which their current assets minus total debt are more than their market capitalization. This value analysis can be done systematically, with minimal human intervention, using heuristics and eventually automating it. Some even consider Ben Graham as Quant ⁴.These examples show how the quantitative approach is applicable across investment philosophies.

There is also the possibility of making joint decision making between quantitative (Quants) and fundamental methods, also called Quantamental. And although the term is a long one, a boom in alternative data analysis has been generated to support this alliance.

It's not just efficiency
Among the benefits of having a quantitative strategy is clearly the ability to use a lot of data, or being able to use a lot of data, or unconventional data, to make decisions. From monitoring a business by its web searches, to the economy and inflation by web prices. The possibilities are limited by the imagination and needs of the analyst.

Quantitative analysis has different benefits and risks than traditional fundamental analysis. While a priori this is neither better nor worse in itself, it is a good way to diversify the portfolio.

Like decision makers, all models have biases of their own. The problem can be exacerbated when these are not identified and mitigated, especially if they are leveraged. Such was the case with the Long-Term Capital Management fund, where its seemingly stellar team failed to understand the risks of its model and, thanks to the exorbitant level of debt, almost took economy ⁵ down with it ^.

Beware of fooling yourself 

Designing a new model is more of an art than a science, and is a combination of financial theory, statistics, computer science and other disciplines. is a combination of financial theory, statistics, computer science and other disciplines. But this process can also be fraught with malpractice, whether intentional or unconscious.

"Most people use statistics like a statistics like a drunk to a post, more for support than enlightenment" - Andrew Lang.

The first, implicit by definition in LLM models, is overfitting models. Overfitting parameters to accommodate past observations is unlikely to work with future observations, since the model is not free to exclude observational error as random. It is like someone who always has the logical and detailed answer to something that has already happened, but cannot foresee something before it happens.

The second bad practice is known as Model Mining. In short, it is to test many models, regardless of the theoretical framework, and stick with the one that gave good results in the past, without differentiating whether it was by chance or by the model. Or trying many explanations of a past event and keeping (and repeating) the one that was right.

When making a quantitative model you have to be innovative and look for your team's competitive advantage where others have not ventured. Ideally design the model in a different way than what is available in the literature, because these already published strategies are generally being heavily exploited and have little room to generate value for more investors.

The process of designing, testing and implementing quantitative models requires an open mind. While one can abstract as an academic exercise, reality is always more complex than the theory and the model. Therefore, it is it is essential to be open-minded and look at the model and the process with a critical eye, identifying its shortcomings, and seek to structure it in an anti-fragile way.

A new quantitative strategy, compared to a new fundamental strategy, can be easy to test consistently. You can test how the model would have performed in the past (back testing), and how the investment would have performed retrospectively. Then, if the results are positive, one can replicate the decision making that yielded those results going forward with the same rules considered. And it is at this point that one sees the results of best practices.

Thanks to the exponential growth and access to computing, many quantitative investment strategies exist today, but their potential is still inestimable. Markets are moving in this direction. This is why both analysts and their investors need to understand the enormous potential and risks of a quantitative strategy, and seek to follow best practices so as not to take unnecessary risks in joining this trend.

By Joaquin Guenim Alé, CFA

¹ "The Little Book That Beats the Market", Joel Greenblatt
² "The Intelligent Investor", Benjamin Graham
³ "Security Analysis", Benjamin Graham
⁴ "Ben Graham Was a Quant", Steven P. Greiner
⁵ "When Genius Failed", Roger Lowenstein