High Frequency Trading

The speed of communication is wondrous to behold
 
I guess this is the new buzz word on the street these days, after "recession", "financial breakdown" et al especially after Goldman Sachs piled on millions of dollars using it early this year. To capture the essence of this term in brief, lets say high frequency trading is all to do with speed. To me, it seems to have added a new dimension, a rather more domineering one, to the equation of making money- The faster you can trade, the more money you can make. Can it get any simpler to comprehend? With the advent of computers around 1980, the notion of algorithmic trading started taking shape. As explained in my previous post on "Market Makers" , these algorithmic traders started assuming the roles of market makers too. With this new role, came a few added provisions and thus new ways to make money. These set of algorithmic traders became distinguished in their approach of trading and gave rise to what we now know as - High Frequency Trading (HFT). In this, these traders take advantage of their favourable position  (exchange rebates, high speed connectivity to market and so on) to craft new trading strategies in which the basis is to buy and sell at very small intervals, very quickly! This new strategy was reinforced by the exchanges reducing the tick size (minimum value by which price of stock can move) from 1/16 to 0.01. This allowed HFT traders to quote bid and ask prices with  minimum spreads to increase their profits.
A few of the strategies used by these HFT traders are as follows:-

1. The first one involves exploiting the rebates given by the exchange to these market makers (aka HFT traders) on each transaction due to their "noble deed" of providing liquidity to the market. Suppose a institutional trader wants to buy 100 stocks of IBM at $20. He is using algorithm engines for firing his orders into the market. These algorithms often slice the orders while releasing them ot market using strategies like VWAP, TWAP etc to get the best deal. Now, while they are doing this, a HFT trading program is sniffing on these orders. Lets say, the institutional trader got  fill for 10 orders at $20. Then later, he got a fill for another 15 at $20. using this data, the HFT program can track that these orders are coming from an algorithmic engine and quickly place a bid on 75 IBM at @20.01. This will cause the HFT program to get the next fills instead of the algorithmic engine because HFT is willing to buy at a higher price. Now, since HFT  knows that the algo engine is waiting fill its remaining 75 IBM at $20, it will sell it 75 IBM at $20.01 (yes, it didnt try to hole a spread here, indeed). Now, what we see is, the HFT program simply bought IBM and sold to algo engine at same price without making any profit on spread. We call these dead transactions. But even on these dead transactions, the exchange gives market makers (HFT's) a rebate of 1/4th penny. So, for two dead transactions, HFT got a 1/2 penny profit while the algo engine was made to pay a penny more. This 1/2 penny scaled up to million dead transactions will be quite some huh.
2. The second one is what is called predatory algorithms. In simple terms, think of these as algorithms which "feed" on other more innocent algorithms. Typically, a tolerable  range of execution prices is fed into algorithmic engines rather than  one stringent limiting value. These engine then post bid/ask prices in increments of ticker. So if one engine bids for 21.1 and second bids for 21.2, then first one  raises its bid to 21.3 until it reaches its limit. The predatory algorithms take advantage of this behaviour. In our e.g., lets say the institutional traders algorithm has a range of $20 - $20.9. So when it makes few initial bids and gets fills, the HFT program sniffs on it as before and tracks it down. Now, next step is to post competitive values against the  algo engine. So HFT program  will bid for $20.1. Seeing this, the algo engine bids for $20.2 and this continues till the HFT program stretches the algo engine bid till its limit (or close to it). At that point, it withdraws its bid, and instead goes ahead and sells short to the  algo engine at $20.9. Now since the HFT program knows that this new high value is just artificial and is soon going to come down, it will wait until it does. Once prices come down, HFT program will buy  from market at low price and make delivery to the algo engine for  the short sell it made before. And the net spread is pocketed by the HFT program.
3. In the third strategy, the HFT program makes use of the fact that its been allowed to act as a market maker. In this, like before, institutional trader wants to buy at $20. however, his limit is say till $20.9. like before, the HFT will track the algo engine down. Once that is done, its job now is to find the limiting price of the algo engine. For this, it will ping to the algo engine. It does this by sending small namesake orders at varying prices. It will first send an order of say 1 lot at $21.1. Nothing happens so it cancels the order and places another at $21. Again nothing happens, so it cancels it and places another at $20.9. At this, the algo engine will hit a trade with HFT. So now, HFT knows the limiting value of the algo engine. It then immediately makes a bid a tick higher than original bid of algo engine i.e. it will make a bid for $20.1 and pick up all the shares from market. It then wastes no time before it posts an  ask price of $20.9 on its 100 shares, which in high  probability will be bought by the  algo engine. Thus, as before, HFT pocketed the spread.
4. The forth strategy is not so much to do with using complex algorithms but more to do with sheer common sense. The exchange typically rents some space in the same location as its own, to algorithmic traders to place their large server racks and operate from there. In return they get a handsome rent for it. Now because of being located in the same location as the exchange, these algorithms get faster connectivity to the market due to shorter length of cables connecting them to the market (!!! honestly :-D). This reduces the latency of information transfer, and these engine thus get market quotes a few hundredths of second faster than others. This is known as co-location. Least to say, they are the first ones to act on the market and make money.