A losing day once every 8 years...

March 20th, 2014, 06:34 PM

Interesting theoretical calcs on how a market make might make consistent profits with a 52.5% win rate and w:l of 1:1. Original link here.

Quoting

A simple model for the P&L of a Market Maker (how many losing days in a year?)

Say that there's a probability p of making one unit of P&L (e.g. 1 tick = 1 cent) and a probability 1-p of losing the same amount of money
Draw n samples of a Bernoulli distribution with probability p, subtract 1 and multiply by 2 to get a coin toss mapped into the P&L distribution above
Let's have n equal to something like 7*60*x, where we have x trades every minute over 7 hours of trading
We then accumulate the P&L (assuming no further risk measures that would stop trading if the accrued loss was too high) for each run.
We look at the distribution of accumulated P&L to estimate how many losing days such a trader would have

If we have p equal to approximately 52.5%, and 10 trades per minute, it looks like this trader would have one losing day every 8 years.

Find a small edge and exploit it repeatedly.

March 21st, 2014, 10:38 AM

This is the goal of institutional algorithms - to essentially profit many times from small movements. In practice you need to watch the algo closely for this sort of work it is easy for it to go wrong as you're working on the margins, and that makes it expensive, hence the reason its currently the domain of the big boys, although I do concede retail investors can to HFT its much harder and more time intensive than it might first seem. An algo manager I know tells me his firm is constantly updating their code to not only keep up with the market diversity but also to keep fighting the other algo's as these things are learning and trying to exploit their competing algo's in the market (hence the now wide use of spoofing out there).

Interesting topic!

A losing day once every 8 years...

Discussion in Psychology and Money Management

A losing day once every 8 years...