Please Note: Blog posts are not selected, edited or screened by Seeking Alpha editors.

High Frequency Trading -- Results from Simulation


Defenders of High Frequency Trading (HFT) are fond of asserting that program trading increases liquidity and can be shown to narrow the bid/ask spread.  This would seem to be a good thing for retail and institutional investors.  However, as I will argue here, this may very well not be the case.  I've run computer simulations in regards to HFT strategies and the results are revealing.  A nice primer of how High Frequency Trading works can be found in the Themis white paper,  For those not interested in reading the entire paper, below in two paragraphs is how it works.  Then, I will show you what my simulations found. 

Most investors are at least passingly aware that the majority of trades on the NASDAQ, NYSE and NYMEX are, in fact, program trades.  Rather than someone deciding to buy some stock and placing an order, a computer program fed with market data makes the buy and sell decisions itself.  What you may not know is that high-frequency/volume trading accounts for 70% of trading volume (Themis et. al.).  Recently over 50% of program trading on the NYSE is now done by Goldman Sachs and 90% of that is for their own proprietary trading (i.e. their own account.)  (NYSE public figures as reported by Zero Hedge).  Stop and think about that.  If these figures are right, 30-35% of all the trading on the NYSE is Goldman Sachs for their own benefit. 

High Frequency Trading (HFT) is computer trading that buys and sells large number of shares in a very short period of time.  At it's core is the realization that if I (my program, that is) can trade faster than you and you are a Large Institutional investor (NYSE:LII) with a large order, I can scalp a few cents off of each share you buy or sell.  To do this I first try to recognize that you are a LII with a large order by recognizing the pattern of buys (or sells) issued by the trading programs you use.  If, for example, I see a telltale series of purchases of 100 to 500 share sales hitting the tape I will offer a series of rising bids to determine how far I can bid you up, or post a series of descending asks to see where you will start buying.  When I've determined the range you're willing to pay, then, because I'm faster than you, I hoover up any and all shares offered in the market at less than the top of your range and sell them back to you right at the top of your range.  Alternatively, I can put out a bid one penny below the top of your range to keep you on your highest bid and then short shares to you at that bid.  When you're done I'll remove my bid, let the bid fall back and cover at a lower price.  Simulation shows this works iff:

1) I can trade faster than you, 
2) my cost of trading is 0 (or negative as it is for a supplemental liquidity provider), and 
3) I can determine that you are an LII with a large order and figure out your range.  

In college, for fun, some of us geeks would write computer programs that would battle each other in a virtual market place. At each turn, the programs would try to out think the others.  I sometimes wonder if any of  my classmates ended up programming for investment banks.  Note that being able to trade faster and for zero cost translates into a sustainable ability to allocate cash to yourself.   To the extent that LII programs can be made smarter or join the club and thereby neutralize one or more of these advantages, they can fight back and should.

What does this all mean?

Assuming that the description of HFTs in the Themis paper is correct, there are some interesting results of rampant program trading.  Generally, over some arbitrary period of time, absent of program trading, trades are distributed between the bid and ask prices as below.
In a rising market the distribution will be skewed to the ask, in a falling market, toward the bid.  A narrow bid/ask spread is considered efficient for the market and good for retail investors as it makes the distribution tall and skinny: more people get a similar price closer to the average market price.  However this is not the whole story.

The effect of HFT program trading as described above is to force LII buyers towards the ask, and LII sellers toward the bid.   The black line curve below.

The price differential between the trades in the red shaded area and the trades in the blue shaded area roughly corresponds to profit allocated to the HFT programs.  Thus citing a tight bid/ask spread at a single instant as a benefit of program trading does not tell the whole story.  It's a tighter bid, but now where you would want it or where it would otherwise be.  Additionally the bid/ask that you see is not for all intents and purposes the bid ask that the HFT programs sees.  Creating this artificial bid/ask for the LII is, of course, the whole idea.

There is another type of program trading that we may briefly consider, namely strategies that seek to move the market past certain price points where the program believes that other programs or individuals wait to buy or sell.  This may include holding an option simultaneously.  The most obvious example is shorting enough stock in nervous times to trigger panic selling.  This strategy can also work at breakout points, over-bought/sold points and other technical patterns. My simulations show that as program trading increases, the bid/ask spread tightens at the expense of short term volatility which rises as programs race each other up and down and volume rises.  Bid/Ask spreads are computed as of a moment, but other things being equal, increased volatility means a wider effective bid/ask spread over a period of time as actual trades are distributed over a larger range.  To non-program investors it is effectively a tax on investing.  

Final Thoughts

The first example of computer crime I heard about in school was the case of a programmer for a large bank who noticed that interest calculations were being truncated at the penny.  It would have been more accurate to round up or down to the nearest cent but none-the-less the bank truncated. Thus with every interest calculation a small amount of money was effectively vaporized.  The programmer modified the program to transfer the fractional cents to an account controlled by him.  There were enough transactions that this amounted to a lot of money.  Amazingly, though he received new bracelets for it, neither the bank nor its customers were in any way worse off, as each received exactly the same amount as before.  Even the vaporized liquidity was returned to the system, if oddly allocated. The same cannot be said of the HFT programs described here and the firms which employ them.

I was taught as a child that the purpose of the stock market is to spread the risk of business ventures as well as to spread the wealth created by their success.  I hold to that today.  That purpose is met when entrepreneurs start more businesses, more people are employed, higher wages are paid, and wealth is created.  High speed program trading of the kind described here does none of this.  It does not appear to meaningfully increase liquidity, nor meaningfully reduce the bid/ask ratio.  It is effectively a tax on investment trading and therefore antithetical to the purpose of the market as it allocates sizable returns to the unproductive.  Strangely the fastest and strongest actions of the Justice Department and SEC in recent days have been not to protect investors from HFT programs but instead to support and perpetuate their use. As investors and citizens we may reluctantly concede that the use of HFT and other programs is perfectly legal under current rules, but concomitantly conclude they are of no value to the market and should be eliminated.