Monday, 7 April 2014

HFT - two choices: making money every day or oblivion

If an HFT makes money every day the system must be rigged!

It's not rigged.  It's just math!

I see since Flash Boys came out, HFTs have been copping flak for making money consistently.

Virtu had one down day in 1200 or so days. It must be rigged.

Well, the simple fact is that if you don't make money every day, then you're doing it wrong!

You do, or you die. It's that simple.

HFT implies that you do a lot of trades. If you do a lot of trades with a smidge of favourability, you should always be winning. It's just simple mathematics. The law of big numbers. A systematic screw-up for a day may cause an odd loss but that is a different risk category. You seriously wouldn't be doing a zillion trades of anything without an overall positive expectation. The law of big numbers also works against you. If you're on the wrong side of the fragile line you're practically guaranteed to lose.

This is not to say that you have a "cooked" system where every trade is a winner.

Let's look at some numbers:

Say you have a trade that makes you $0.01 per trade when you get it right and you lose $0.011 per trade when you get it wrong.

Let's now assume you do 10,000 trades a day.

What winning trade % do you need to win at for almost every day to be a winner?

Let's look:
  • 51% means roughly 99.7% of days are losers, a bit over 0.2% of days are winners
You probably wouldn't be happy with that.
  • 53% means 89% of days are winners
  • 54% means 99.9% of days are winners - you lose one day in four years
  • 55% means you really shouldn't lose
The percentages are tight. It is a matter of getting the profit / loss distribution for your trade outcomes right so a lot of trades make it hard to lose money.

Fundamentally, if you're a high frequency trader, you're doing a lot of trades. That's pretty much the definition of high frequency, otherwise you're a low frequency trader. A lot of trades are silly if you don't have a positive expectation. If you have a positive expectation then a lot of trades means you really should have a very high chance of winning on every day even if each trade's win chance is a bit like the toss of a coin.

Let's look at the same situation if you only did 100 trades a day. A hundred is quite a few but not really high frequency.
  • 51% trade win probably = about 38% of days are winners
  • 54% = about 62% of days are winners
  • 55% = about 69% of days are winners
Not quite as good as having a lot of trades at 55%. However at 51% you were previously pretty much guaranteed to lose with lots of trades but you may live a few days longer with only 100 trades per day. This is also why if you want to bet on red or black at roulette, where the house has an edge thanks to zero, you should just make one large single bet to maximise your chances of not losing.

You can see that lots of trades help improve the guarantees. You win or lose more consistently.

The hard thing about HFT is that if it was easy, everyone would do it. There are no easy trades with the level of sophistication that exists in mature markets. It is challenging and you are constantly being pushed to the margins as someone else is always prepared to make a little less and take all your edge if they can. HFT market makers fight compulsively over the scraps and win by making less than another team. It's a bare knuckle fight against other market makers where your reward is to make less money than your opponent if you win. You in turn get beaten by someone prepared to make even less. You are scared of large trades or better informed traders that will trade through you and screw with your distribution of returns. There is little room for error.

It is simple. An HFT should be making money every day or they will be going out of business. There is money or there is oblivion. There is never any forgiveness, just paranoia.  HFTs are terrified of adverse selection. Worse still, other market makers are trying desperately to replace you by being a fraction better. Lots of smart firms are always shutting down in HFT land. It is a tough business.

So, what's up with traders making money every day?

It's not rigged.  It's just math.


PS: Survivorship bias for HFT types means that all should have consistent profits if they remain in business for an extended period of time. I find that an interesting thought.

Earlier description of how HFT market makers make money
Earlier reasoning as to why investors and speculators can both win in markets


Footnote:  I spent quite a few years as a positioning style trader with a global portfolio of currency, short and long interest rates, equity indices, and some commodities.  I made money every year but it was a battle.  Each day was a 48% chance of a win. So, overall I was usually wrong. I was successful because the average winning day was 1.6 times the average losing day. It was pretty scary though, just missing your dozen best days in a year would be the difference between an annual profit or an annual loss. It was also quite funny when people asked for a comment on the markets as it was hard to explain that with my trading I was more likely to be wrong than right for any given day.

HFT is pretty similar. It is also just a game of profit and loss distribution. It is a lot easier on the psychology. Instead of waiting for a whole year to see if things work out, you should know at the end of each day with a large enough number of trades. Even worse for the psychology is that as a position style trader normally the profits are positively autocorrelated which means that if you are losing you expect to continue losing. Worse, you can't do anything about because if you over trade, resulting in extra frictional costs, you're only guaranteeing an eventual loss. HFT has better daily psychology but higher paranoia as there is a lot that can go wrong.


Footnote: This is vastly simplified but correct in essence.  Profits and loss tend to be positively autocorrelated which makes things more volatile in practice. Distributions are nowhere near as simple which is why there are PhDs running around with computers attached to their fingers.


  1. I see a smarter and more informed person than I has put out the same point of view but with real data and much more coherently:

    Institutional Investor: A much needed HFT primer for 'Flash Boys' author Michael Lewis

  2. So HFT firms are more about optimizing trades in accordance with profit/loss distributions rather than price prediction?

  3. Price prediction, even quite noisy price prediction, may help optimise your profit/loss distribution by biasing towards more appropriate trades. Do you need it? No. Why do it? In a finely balanced game, every little bit helps. It's all about the profit/loss distribution.