High Frequency Computer Trading: How Does It Work?

High Frequency Trading did not start with the Flash Crash on May 6th, but the crash made High Frequency Trading HFT more visible.

Major investment banks and hedge funds have been building computer centers within yards of Wall Street computer center to gain milliseconds of advantage in rapid algorithmically-driven trades. How does this work?

I wanted to know whether HFT was just computerization of what floor traders have always done, or it is some new kind of scam. Second, the financial press always says HFT techniques are "too complicated to understand"  similar to derivatives trading. In time I have learned that derivatives trading is not really that hard, and so I set out to learn ways that HFT makes money.

Automated trades may account for 80% of total stock trading volume. This is somewhat controversial because to get a number this high journalists include all computerized trading schemes for whatever purpose. Some of these are long terms trading schemes and not high frequency.

Trading houses are executing trades for a few hundreths of a cent per share. This is far less than retail investors pay, and it makes makes it economical to go after small opportunities in the market. In this sense the high trade volumes were see are caused by the low price of trading, which only makes sense.

The following are ways to make money using high frequency trading techniques.

Ten Ways to Make Money using HFT

1. Suppose someone is trading a large block of stock in pieces to hide it. One can analyze lots of trades to find a pattern of sales that are the sign of a large transaction. The program then buys or sells to capitalize on the trade, that is to copy the original trade.. This disadvantages the original trader, but is no different than what a floor trader would have done.  There is an advantage of having fast access to the market, and it is parasitic on the knowledge of the original investor. This HFT hurts liquidity because it makes each trade more extreme.

2. Placing trades where one party is both buyer and seller to manipulate the price. When one gets the price where one wants it, then place a bid trade to profit from it. This is illegal.

3. Like #2, but not actually making the trade -- this is legal. One offers to make trades and then withdraws the offer milliseconds later before anyone can respond. If one floods the system with zillions of trades that don't execute, it confuses competitors. It will take them time to figure out that I am masking something, and in the meantime I can concentrate on the real trading. That is, I can execute a big trade or analyze the other trades for information without distraction.

4. Like #3, we place tiny trades in an attempt to determine the elasticity of demand; that is how much is the market willing to pay; then we use that knowledge in our future trading. Alternatively, the tiny trades may confuse simple-minded competitors ( who only look at numbers of trades and not dollars) into thinking the market is trending on-way or the other.

5. Market Making - I place an offer to sell above the current price, and a bid to buy below the current price. Market volatility and the illiquidity of the market will ensure I get some orders, and I benefit from the bid-ask spread.   This is the classic high frequency trading play.

According to Wikipedia, ATD Automated Trading Desk owned by Citigroup does this, and accounts for 5% of all trades. Kansas City based TradeBot accounts for another 5%.

Tradebot, ATD and competitor Getco are more than schemes to make money. They are mini-stock exchanges that allows brokerages to save money on zillions of retail trades. The bid-ask spread is a measure of the inefficiency of the market, and operations like ATD make that spread smaller -- thus making the market more efficient. Nothing the matter with that. [Tradebot also has more speculative operations, besides only this.]

6. Technical Analysis = Quantitiative Analysis = Statistical Trend Analysis - Some would argue that technical analysis is not high frequency trading because it involves holding positions over a series of days. Others argue that its principles can be applied within a day. Technical analysis involves predicting future price trends from past price trends using fairly straight-forward statistics on the price and volume of trading. I think technical analysis is worthless, so I find it hard to explain how it works, i.e. it doesn't work.

7. Statistical Arbitrage - Assume that over the years, a group of stocks A, B and C always moves together - not exactly together, but that their movements are correlated. If there is a news event or investment reason that independent investors buy stock A, there is a significant chance the same factors will affect B and C, so the high frequency trading program buys B and C too.

This assumes that the correlations between A, B and C are real enough to persist into today's trading, and sometimes it does not, but on average it does.

Manoj Narang says, "Because there are far fewer systematic drivers than there are securities which depend on them, correlation between securities is guaranteed to exist!"  This makes a lot of sense.

One would have thought that finding correlations between news events and stock prices required a human, but the machines can automatically trade on some average stock correlation and get a short term gain. A human analyst could do a better job of linking a particular news item, say a weather report, to those stocks dependent on it, but the high frequency trader would have already gotten his/her cut. When the human investors arrive, the HFT program exits.

Statistical arbitrage is probably the best most legitimate HFT strategy. It takes knowledge from the market and combines it with additional analysis to make short term profitable trades. When investment banks talk about "Black Box" secret trading programs, I think this is what they mean.

So far I have been discussing statistical arbitrage applied to stocks, but one can use it for bonds, commodities, currencies, or futures. One can use it synchronize results between exchanges or interest rates between countries.

8. Intermarket Sweep Orders - These allow some traders to have priority over other traders. If there are more buy offers at price A, than there are sell offers at price A, some will get executed and others won't.  Intermarket Sweep Orders ISOs get an earlier place in line and get executed before others. Traders like me without access to ISO's pay more.  Smaller investors pay an average of $0.013 more per share when buying for this reason. 

9. Latency Trading - The same idea as above where the HFT trader gets trades executed faster simply because they have high speed connections and are physically close to the exchange. I get some priority because I am nearby. If an event happens, I get an advantage because I am closer to the market. This is just a computerized version of proximity advantages that have always existed.

If I am reacting to trades on the market, then also get an advantage in learning about these trades if I am nearby. In-coming knowledge is a second kind of latency advantage.

10. Rebate Capture - Some exchanges offer rebates to brokerage firms who place trades with them. Some firms will place trades on both buy and sell sides to generate fake volume to earn rebate income from the exchanges. Obviously this is a sleazy practice that adds nothing to economic value. I doubt one could actually make money doing this because the rebates are small between 0.01-0.25 cents per share depending on the stock and the market. NASDAQ rebates are higher; NYSE rebates are lower. Rates are structured so big traders get more generous rebate rates.


Closing thoughts

On the May 6, 2010 Flash Crash, the high frequency traders all made tons of money because they profit on volatility. They argue -- and I tend to agree -- that their systems stabilize the market since they win by reducing fluctuations.

High frequency traders keep their shares less than a day, perhaps an average of ten minutes. They argue that they simply cannot influence the daily price of a stock because they have executed both sides of the transaction during that day.

Who caused the Flashcrash? It was not one of the big-guy, high-frequency traders, but instead an interaction between automated investment programs with access to a substantial amount of stock. Whoever it was is probably out-of-work now, and whoever lost money on it is probably trying to keep quiet.