This quote is not mine, nor is it part of my field. The author was George E. P. Box, a British statistician, who wanted to emphasize that statistical models seek to achieve a level of precision regarding human behavior that is (currently) impossible to achieve. However, alluding to the second part of the quote, Box states that, although these models are not exact, they can be quite useful in predicting general behaviors of the study in question.
Following another similar analysis of models, in “Essays on Positive Economics,” Milton Friedman, Nobel Prize winner in Economics in 1976, tells us that the most important thing about an economic model is not its content nor its logical structure, but its predictive capacity. He believes that the best models are those that predict most effectively, without focusing so heavily on their components.
How can we apply this to the world of Investment Banking?
In the corporate world we tend to use different models for different industries and sectors. The model of a digital marketing agency will not be even remotely similar to that of a construction company and the latter will not be similar to that of a consumer goods company either.
Each model will have a different objective, but in general, they are used to understand the past and predict the dynamics of the business in question (the latter being very important since these models evolve over time) and how it might yield different results when facing internal or external variations or shocks.
Now, how relevant is detail in a financial model?
When we do a valuation using DCF (Discounted Cash Flow, the valuation style with the greatest consensus in the world), what matters is the generation of cash, so we can then discount it in the present time and determine how much the company we are analyzing could be worth.
It is essential to understand that you cannot create an overly simplistic model that effortlessly projects revenue, expenses, working capital and capex to arrive at a number you more or less like. It’s important to pay close attention to the most important business lines and the valuation.
At the same time, it can be counterproductive to have every last detail modeled. At first, it sounds fantastic to have absolutely everything under control, but it’s often time-consuming and you probably won’t have enough time due to circumstances in your workplace. For example, “I need the valuation within a month because we have an offer from our client.” In a situation like this, you can’t waste time. Furthermore, it’s clearly more difficult to manage in the event of a shock that completely affects business plans (the new US tariff policy is a good example). Many companies will surely have to make decisions regarding this and an overly complex model may not help them understand the effects of the shock due to its structure and the tedious nature of changing all the variables again. In short, they end up creating a new model because it’s easier than changing the one they already have.
Taking all of this into consideration, I don’t mean to say that everything should be simplified to the max: far from it. Personally, I’m quite a fan of managing many variables simultaneously, but I know their limitations. After all, the goal of models is to predict in order to make the best decisions. Therefore, it’s not advisable to have an overly simplified model or one that’s too detailed. The ideal scenario is to have a model (in this case, a financial one) that reflects the most relevant business flows, that is adaptable to future changes (both external and internal) and that predicts effectively.
I always like to keep in mind a concept I learned in statistics called the “Parsimony Principle,” which states that, when faced with two models that predict with the same accuracy, one should stick with the simpler one.
