Sunday, July 14, 2013

Beware Discrete Auctions

There's an interesting paper on High Frequency Trading (HFT) by Budish, Cramton, and Shim from the U of Chicago. They set up an interesting model, but then propose at the end that batching eliminates the HFT arms race, both because it reduces the value of tiny speed advantages and because it transforms competition on speed into competition on price. I think that solution is interesting, but would prefer the following

  1. add a 20ish millisecond randomizer to any incoming order (ie, a lag of anywhere from 0 to 20 milliseconds). Thus, if you know you are getting randomized, the marginal value of 1 millisecond is much smaller to the innovator, your place in the queue is noisy. 
  2. make a rule that those who receive fees (liquidity providers) must have such orders exist for at least 1 second. That would get rid of a lot of flash quotes that are trying to be too clever and simply clog the bandwidth.

In any case, we have several different exchanges, and so if there's a simple way to increase welfare, an exchange that adds any such rule would generate more traffic, and ultimately, imitation. Yet, that's how paying for liquidity provision (passive quotes) originated, as well as a bunch of other tactics. One thinks they are annoying features created in a smoke-filled room, but instead they were simply the result of giving market participants what they want. A retail trader can always opt-out by simply trading at the Volume-Weighted-Average Price (VWAP), because that gives you the average daily price that averages out all these shenanigans, so I really don't have any sympathy for those who are bothered by it: if you don't like the intraday game, don't play it!

But, I was intrigued by the idea of batch auctions, because I've heard people who think discrete auctions are better than continuous ones. I think they are not very good, and a simple example is an extreme of this, one can look at the performance of closing vs. opening prices. One is at the end of continuous trading, one where there's no trading. What happens?

Well, I looked at over 1000 stocks, from 2000 to today.  I excluded really small, low-priced stocks.   The Open-Open returns have slightly higher volatility (2% higher), but more importantly, there's a lot more 'mean reversion'.  The graph below shows the future returns(O-O or C-C), sorted each day into deciles by the immediate prior return (O-O or C-C, respectively), then averaged over all those days.

Basically, markets open at extremes that are quickly erased, allowing the market makers to pocket nice premiums for these temporary imbalances (you can't make money off this if you aren't a specialist).  Continuous trading takes such trades away from monopolists, and allows competition to work.  

These authors propose more frequent auctions to be sure, but the logic remains.  


JoshK said...

There have been many attempts to create intra-day auctions and they have failed. I'm sure no one remembers NYSE Matchpoint any more.

Mercury said...

I like your two recommendations. I’ve thought of continuous batch auctions before as a possible solution but now I see the light that a time stamp randomizer within a one second (or whatever) window is really the best work-around for the “arms race” of access speed. It’s the only way to get around the fact that nodes physically closer to the exchange will otherwise always have an advantage. The time between auctions just has to be long enough to be a generous amount of time for someone on the other side of the world to sling an order to the exchange.

The second one has been brought up by many others in one form or another (fee for cancelled or multiple cancelled orders, minimum display time etc.) but it’s still a good idea. Quote stuffing is BS and it needs to be made impossible or prohibitively expensive, period.

But I still need to be talked out of my belief that one, centralized marketplace is more conducive to transparency and price discovery (more or less by definition) than a system biased toward fragmentation in the name of competition - even when virtually aggregated for market participants. Inevitably, giving market participants what they want –which sounds so ultra-free market- includes exploitable advantages that some are better able to exploit than others with sheer firepower.

Fish Goldstein said...

Our U Chicago Colleagues are delusional: They are trying to solve a problem that doesn't exist.

People just need to stop using "market" orders. Would you buy a house on the "market" price, by telling your Realtor that you'll agree to ANY price the seller wants? Probably not. And there's no reason to do that with stocks either.

Of course, one reason to use market trades is because the trader is too lazy to figure out limits. In that case, the trader pays for that convenience.

Regarding "clogging bandwidth" there is no evidence that flash orders are causing dissarray in the exchange. Even if they were, that would just be a technical issue of the exchange, and should be handled like any DOS attack.

Third, their solution won't work. Batching by 1 second would lead to people timing the system, and figuring out how do direct orders into one auction or another to optimize trading. It would actually add complexity into the market that would give an advantage to the HFTs. The authors should try their proposed solution at Kovler Cafe, and see if it makes getting lunch more orderly.