The 80/20 rule

How I make more money trading whilst doing less work

Once upon a time, I stepped onto a trading floor for the first time, as an eager trading intern, I had no idea what I’d got myself into.

I had next-to-0 knowledge about trading, but somehow I found myself in the middle of an index options market making desk, shaking hands & meeting the team. To give you an idea of how little I knew, I didn’t have a brokerage account, I had never owned any stocks, I didn’t know what longing or shorting meant. Yet here I was¹. I had expected to walk into some kind of computer lab with rows of computer nerds tuning complex models, instead, I was met with walls of screens and traders yelling numbers & greek letters at each other. It was here that I was first introduced to the 80/20 rule, a rule that has guided me for my whole career in trading since.

For those that don’t know it, the 80/20 rule, a.k.a the pareto principle is loosely defined as 80% of outcomes come from 20% of causes. We extend this to a work ethic to say 80% of the solution can be found with 20% of the work. Crucially, this implies that the remaining 20% of the solution will take 80% of the work, or 4 times more effort². I hope you’re following me so far, if you are, you might be asking “what has this got to do with trading though?”.

Great question.

To answer it, I’ll take you back to how I had this hammered into me as a QR³ early in my career. The senior trader, let’s call him Tim⁴, who was responsible for wielding my autism would often see me starting a complex task, perhaps I’d have read a paper on a state of the art model from another industry & I wanted to try use it for forecasting volatility clusters. Upon seeing this, he’d tell me “but have you tried a linear regression first.” to which I would invariably answer “no” and try explain heteroskedasticity and say a lot of big words. Tim would tell me “80/20 it”, and tell me he didn’t really care if it was that accurate, he just wanted a rough idea of how many times to click the button to move the vol curve up or down. Regardless, I would go back and try a simple regression first, initially I complied in an effort to prove how much better my method would be. However, after several repeats of this, I realised that I would spend hours and sometimes days on something shinier and more complex, only for it to not work, or to overfit 4 times out of 5, a pretty shit hit rate & if we frame it as a percent, I had a 20% hit rate...

Over time I learnt that if I just did a quick, sane, regression & took it back to Tim, this was far more useful for trading & yielded far more PnL improvements for our desk. I was able to tackle far more problems than if I was to try the hard way every time. If I can do a roughly 80% as good of a job but do that on 4 times as many problems on the never ending list of possible improvements, you actually end up with more money⁵.

What matters in trading isn’t that you’re right, what matters, is that you make money. And to make money, you have to be competitive. Being competitive means being fast and adapting, and I don’t mean fast in a HFT sense, but in a time-to-market sense. Generally with any alpha or opportunity, there will be a spectrum of all possible opportunities, with some of the trades being easier and more profitable than others.

A great example is a basis trade, the classic cash and carry. The biggest dislocations are really fucking obvious and they stick around for ages, you just rank the perps by funding rate and take the top one. You don’t need a fancy algo to trade these, you can click some buttons. You can keep doing this for maybe 80% of the opportunities you see. If you wanted to get 100% of the opportunities, then it’s a bit more effort. But of these 80%, you probably captured 95% of the available profit. If you waited until you had a gold plated, diamond encrusted algo system to trade them, then the juicy trades may have already gone (something something, short term speedy, long term greedy).

As my career has gone by and I have met & interviewed other traders, I have seen many a pytorch wielding young quant fall into the same pitfalls as I did. I see it often on twitter too. So if you are to ask me for advice on your quant task & you’re using some random forest or neural net, I will ask you “but have you tried a linear regression first”.

1. If you would like to hear about how I ended up in trading and maybe advice(?) for others wishing to make a career of it, let me know via a comment.

2. One important exception to the 80/20 rule is when evaluating probabilities, then it is 50/50. I learnt this one from my good friend Worst Contrarian aka larp capital

3. Quant Researcher for the uninitiated

4. Tim is obviously not his real name

5. If not from trading, you at the very least get glowing reviews from your bosses which in turn gets you more salary/bonus dollars