/ A tale of two prediction games

October 3, 2017

marc-rafanell-lopez-342508We recently recommended the book “Everybody Lies: Big Data, New Data, and What the Internet Can Tell Us About Who We Really Are” by @Seth Stephens-Davidowitz. A particular chapter in the book that caught our attention is chapter 7, because it relates to what we do: assessing acquisition opportunities. The chapter tells the story of how Seth and Larry Summers, former President of Harvard University and US Secretary of the Treasury, tried to use data science to predict stock market performance, and failed. We, on our side, use data science to predict private company performance, and we believe that it works.

So, are these two prediction games similar or different?

Seth gives the following explanations for the failure to predict stock market performance:

  1. Huge resources.

The game of stock market predictions attracts enormous resources. A lot of what Big Data can do is already done. Seth makes a point of how much Larry Summers respects hedge funds and their analytical ability. They are smart in their use of big data.

  1. Efficient markets.

Consequently, insights from big data are already embedded in the stock performance. So, the financial markets work efficiently and these insights have no predictive value.

There are so many variables to test (dimensions) that one of them is bound to be statistically significant just by random chance – without causation, correlation or any relationship of predictive value.

  1. Overemphasis on what is measurable.

The way Seth puts it “the things we can measure are often not exactly what we care about”.

The world of small and medium- size private companies is a vastly different world.

  1. Many of the companies we analyze focus on small markets, sometimes national, sometimes global, but often small. More often than not, there are less than a dozen competitors. How many knowledgeable people know these markets? If each company has five top managers who know the market and think about it, this would mean that a few dozen professionals hold the market knowledge. They have first hand, operating experience of the market, but they dedicate very few resources to an analytical approach. So, in term of resources dedicated to data science, it is safe to say that we are at the opposite end of the spectrum compared to hedge funds.
  1. In these markets, judgments on company performance, on market trends and company valuations are often based on opinion. The few dozen market insiders often have different incentives: sometimes they form alliances, and sometimes they are bitter rivals. As a general rule they hide private information, and sometimes deliberately disseminate misleading information. The knowledge is based primarily on a battle of opinions between a limited number of participants. It is not an efficient market place.
  1. We do not suffer from dimensionality. As a general rule, a few fundamental drivers explain the market dynamics. Of course, it is difficult to synthesize them, but this is exactly where data science helps. From there, we build predictions of how the market drivers will move in the future.
  1. Focusing on what is measurable works well for us. This is because what we look for, beyond everything else, is a relationship between the market movements and the company performance. It could be causation, correlation, or any other relationship or proxy with predictive value for the company performance. We may not always be able to explain this relationship, but we work with it, as long it is a good predictor. The question we strive to answer is: if the market moves this way, how would the company perform? For this relationship between market and company to work, and to have predictive value, it is essential that we focus on what is measurable.

Perhaps equally important is what we do not do. We do not try to predict future company valuation; we stop at company performance. We focus on predicting the top line and the EBITDA, but we do not predict future multiples of EBITDA. We see the future company valuation as a judgment formed in the future. In a way, predicting future multiples would be an attempt to form a present judgment on a future judgment. In markets that work akin to a battle of opinions, the level of uncertainly is too high to engage in such a prediction.

But for what we are trying to do, data science, especially tools and analytics designed for Private Equity, are invaluable. They allow us to achieve our main goal, which is to provide an overwhelming intelligence advantage to the potential buyer. This intelligence advantage serves two distinct, but related purposes:

a. reduce deal uncertainty through unique (or at least hard to acquire) knowledge,


b. implement non-obvious post-acquisition growth strategies.


                                                                                                                                                                               Sokrates advisors


This article was originally published as a blog post on our corporate web site: http://www.sokrates-advisors.com