HFT: What is the big differentiator in comparison to other time scales?

Question

High Frequency Trading (HFT) seems to be the big money making mystery machine these days. The purported source of unlimited floods of gelt pouring into the investment shops using it.

For me, HFT is first of all nothing else but trading on a different (i.e. ultra short) time scale. You can make a fast buck, but you can also ruin yourself within seconds. Fast doesn't mean smart, right!

There are not too many players there at the moment so markets are not completely efficient? Fine but the players that are there are the best quants with the finest pieces of technology available on this planet.

My question
What is the biggest differentiator of HFT at the moment (apart form the fact that it is faster)? What can you do here to outperform what you can't do on other time scales? Or is it just like it is with all the other time scales: A few winners and many losers (survivorship bias)? In short: Does quantity turn into quality because it is faster - and if yes, when and why?

Answer 1

HFT seems to be the big money making mystery machine these days.

That's not correct. By its very nature, HFT can only produce a limited amount of revenue. The big money makers are still the large hedge funds that charge 2-and-20 on their \$10B worth of assets.

There are not too many players there at the moment so markets are not completely efficient?

There are tons of players in this space. Feel free to look at any job board.

What is the biggest differentiator of HFT at the moment?

HFT strategies operate at the tick level, and many shops use the full order book data (TotalView, OpenBook etc). That means they can act as market makers, and therefore provide liquidity to the market. There are many equity shops in the US that live off the exchange rebates; search for exchange fee schedule to see the structure at each venue.

That said, it's not easy money. With gross returns often under 1 bps, it's a really tight model and one mistake can seriously injure the firm. There's also the "arms race" to out-do the competitors in speed, which is really tough by itself.

I think we're in a period of over-expansion at the moment and we should expect to see consolidation (either bankruptcies or mergers) in the coming years, just as with any other "hot" industry.

Answer 2

At higher frequencies the coastline is longer. Thus you can be more selective in your entries, or trade more. And by trading more you can get a higher statistical relevance for you system.

When it will stop having an edge, you will be able to stop trading it before it eats into your previous profits. ie: if each day you make 0.5%, in 80 days you will have 40%. Than disaster strikes and you lose 20% in a single day. You are still positive. There were HFT firms posting these kinds of results (but the article didn't mention if they were taking risk or just arbitraging).

You can ruin yourself in seconds without proper risk control, but I don't see how that is different from other types of trading. If you fail to control your algo, you can also fail to control your lower-frequency algo or your human.

More concretely, to quote from http://www.nytimes.com/2010/05/17/business/17trade.html:

The founder of Tradebot, in Kansas City, Mo., told students in 2008 that his firm typically held stocks for 11 seconds. Tradebot, one of the biggest high-frequency traders around, had not had a losing day in four years, he said.

In more recent interviews he was quoted saying that they finally had a losing week :)

Answer 3

"Does quantity turn into quality because it is faster - and if yes, when and why?"

Yes it does. As per The Fundamental Law of Active Management, all else being equal the more positive expectancy bets one can make in a given time period the higher the profitability and return per unit of risk. As Adal responded, statistical significance (or lack) is increased, and depth and length of drawdown is reduced.

Answer 4

I agree with chrisaycock who underlined that they are acting as market-makers.

For a market-marker: reward is bid-ask spread and cost is a combination of volatility $\sigma$ and trend $r$.

Of course a market-marker keeping its inventory $\tau$ seconds, this risk is a combination of $\sigma\sqrt{\tau}$ and trend $r\tau$. So the smaller $\tau$, the smaller the risk.

Of course it is not that simple, because at high frequency, the randomness in the price formation process is no more a diffusion (my formula $\sigma\sqrt{\tau}$ is not that true, you need to use other models, like Hawkes processes, or at least point processes) and the trend is difficult to detect (we are speaking here about micro-trends, but it is not the hardest point since you can unplug your strategy as soon as you are suffering from a trend).