If you ever find yourself having dinner with a bunch of hardcore hedge fund quants, and you want to have a little fun, tell the most successful of the bunch that you heard that he or she is rumored to be a rampant “overfitter.” But suggesting, even in jest, that a hedge fund pro is a heavy overfitter is a good way to disqualify yourself from a Hamptons invitation. That’s because overfitting touches a central nerve within quantitative finance: it’s tough to avoid, even by the most sophisticated of hedge funds.
Taking a step back for a moment, overfitting describes the phenomenon by which a model is made overly complex to explain so-called “idiosyncrasies” in the data under study. That’s an abstruse way of saying that a financial model might look really good in backtesting, but may not have predictive value.
While the concept may be a hot topic within the hedge fund space, it doesn’t come up too much within the broader financial community. However, perhaps the core idea of overfitting should be kept in mind within the context of a professional’s career approach.
Most everyone has a mental system in place — a model of sorts — to make career judgements. Usually, that model is informed by thinking about past successes and failures and assigning a cause for the outcomes: ‘I was promoted ahead of schedule because I performed A, B and C correctly’ and ‘I was axed from another job because I did X, Y and Z poorly.’
The problem with making such assumptions is it’s hard to know if those previous actions or traits have any predictive value. And yet, even sharp executives are apt to make decisions about their future based on a career model that could be built on very tenuous assumptions.
So what, if anything, can be done to improve our career decision-making model? That’s a tough question. But as a start, it might be helpful to remember that many of us look to analyze our own career history so as to fit (or overfit) our official career narrative. Perhaps simply shining a little light on that tendency can help.
To share a career spark, please email Matt Warholak here.