My take on Flash Boys and the HFT problems highlighted in the book
Flash Boys by Michael Lewis is a very well written book about a high frequency finance (HFT) problem that needs to be addressed. Having read the book I wanted to summarize what I learned about the specific problems raised in the book and some possible solutions.
What the book is not?
This book has generated a lot of commentary and criticism in the markets, much of which seem to be related to a mis-understanding as to what this book is about. It is not:
- A general condemnation of electronic trading.
- A general condemnation of algorithms used in electronic trading, whether they are for order routing or predictive algorithms.
What the book is?
A specific example of latency arbitrage where using access to preferential data feeds, co-location of servers and really fast Internet, HFTs can front run large trades and gain a small almost of risk-free return.
How does that work?
It's important to remember that there is no longer just one or two stock markets in downtown New York. Over the last decade the market has split apart into over 40 different exchanges and trading venues that are physically seperate from each other. Given the physical separation, the communications between the venues and customers becomes even more important than it was previously. And this physical separation has introduced a latency in the overall order book of the stock market. Some exchanges and HFT firms have worked together to build a two speed system that provides faster access to some for additional fees.
A very simplified two exchange example helps to illustrate what is going on:
Looking at this example, the consolidated data stream suggests there should be sufficient depth to buy 500 shares offered at 532.87. But here is where the speed comes in as your order doesn't reach both exchanges at the same time. Let's say the order goes to exchange #1 first and is able to buy 300 shares at 532.87. HFTs computers at exchange #1 notices that you have bought 300 shares at 532.87 and races ahead of you to buy the 200 shares on exchange #2 at 532.87 in anticipation of selling them back to you at the new best offer price of 532.88. The end result being that you bought 500 shares at an average price 532.874 instead of 532.87. Seems pretty obvious that if you can be faster than the regular routing algorithms then this will be possible.
Why are they faster than you?
So the key questions is why are they faster than the general market. It appears that a combination of factors leads to HFTs being faster.
- Co-location of servers at the exchanges. Exchanges facilitate this for additional fees.
- Preferential data feeds and access to order flow. Again this creates additional fees for the exchanges.
- Better network infrastructure and design. Some of this is outside of the exchanges and some is preferential treatment of network infrastructure when entering exchange data centers and within them.
- Better design of routing algorithms and building their own consolidated data feeds.
Apart from the last one, all of the advantages are related to fees for additional access that are padding exchange profits but at the expense of creating a two speed system that hurts trust and confidence.
Who does this hurt?
- Only marginally for retail investors doing small transactions that are likely to be executed on a single exchange.
- It hurts large institutions who do big orders that need to be split among exchanges. It can be mitigated to some extent by investing in the necessary infrastructure and human capital to design the more advanced routing and prediction algorithms. But this is a cost on whoever the institutions are ultimately working for (ETFs, mutual funds, pension funds, hedge funds).
What could be done for this specific problem?
A couple of things could help with mitigating some of the problem, although ultimately it can never be totally eliminated as loopholes are always found and gamed.
- Slow the system down a little so that the exchange infrastructure is always faster than any individual market participant.
- Simplify and eliminate many of the complex order types that facilitate the activity of latency arbitrage.
- Don't allow exchanges to create a two speed system by selling preferential treatment to market data or order flow.
What other HFT problems might there be?
While the book isn't specifically about them, there are a couple of other HFT problems are related to latency and should be looked at by the SEC/Attorney General reviews taking place.
What it is: The practice of HFT algorithms pumping out millions of quotes and canceling them almost instantly in order to determine where order flow might be.
Why this should be looked at: It "breaks" individual stocks on a regular basis and whole markets on a periodic basis when market data systems can't keep up with the load. The infrastructure costs required to keep up with this practice are very high and the additional quotes don't provide real liquidity to the markets.
How could it be stopped/reduced: Establish cancellation fees.
Algorithmic News Readers
What it is: News organizations selling preferential access to news feeds that HFTs then use to get a jump on price movements before the news is released to the general public.
Why this should be look at: It's insider trading at the sub-second interval.
How could it be stopped/reduced: Don't allow AP/BusinessWire/Bloomberg/etc. to sell preferential high-speed access to customers before it is released to the public.
A Final Thought
As Michael Lewis always does the book was very well written and entertaining. The most important thing I took from the book was that a two tiered system has been built by the exchanges that creates many opportunities for additional access to information for fees. Given the utility nature and purpose of exchanges it seems like a very bad idea to have allowed this inequity to have grown to the extent that it has. Regulators certainly need to do something to reduce or eliminate the problems to maintain trust and confidence in the markets. At the same time, it is important to remember that electronic and algorithmic trading has been very beneficial overall for the markets by reducing the costs of trading and nothing should be done to reduce this significant benefit.