Exploratory trading

by on December 5, 2012 at 3:16 am in Economics | Permalink

Adam D. Clark-Joseph, from Harvard, has a job market paper on high-frequency trading and here is the abstract:

Using comprehensive, account-labeled message records from the E-mini S&P 500 futures market, I investigate the mechanisms underlying high-frequency traders’ capacity to profitably anticipate price movements. Of the 30 high-frequency traders (HFTs) that I identify in my sample, eight earn positive overall profits on their aggressive orders. I find that all eight of these HFTs consistently lose money on their smallest aggressive orders, and these losses are not explained by inventory management. These losses on small orders, as well as the more-than-offsetting gains on larger orders, could be rationalized if the small orders provided some informational value, and I model how a trader could gather valuable private information by using her own orders in an exploratory manner to learn about market conditions. This co-exploratory trading  model predicts that the market response to the trader’s co-exploratory order would help to explain her earnings on her next order, but would not explain any other traders’ subsequent performance [TC: note that I've tried to correct for garbled text in my reproduction of this last sentence]. In direct confirmation of the model’s predictions, I find that a simple measure of changes in the orderbook immediately following small aggressive orders placed by the eight HFTs explains a significant additional component of those HFTs’ earnings on subsequent, larger orders, but this information offers little or no additional power to explain other traders’ earnings on subsequent orders. These findings help to clarify nature of the information on which HFTs trade and offer a starting point to address the open questions about social welfare implications of high-frequency trading.

Here is a new report of a very different study of HFT, which I have not yet had the chance to read.  On the surface it simply appears to suggest that most people trade too much.

Ray Lopez December 5, 2012 at 3:44 am

From reading this abstract and the NYT article on the other study, I think I can piece together a plausible layman’s explanation of the thesis: HFT (high-freq traders) make money ‘painting the tape’ and ‘front running’. Common illegal(?) practices where in the former you buy and sell simultaneously to yourself to give the appearance of volume, and the latter where you jump in front of another customer if you can deduce. So how is this done legally? That’s for the HFT to know–they trade in milliseconds and have it figured out. I suspect they can ‘guess the price’ what the front-running customer is going to buy/sell at. Or maybe some arbitrage play. IMO a Tobin Tax is called for since the ‘liquidity’ HFT provide dries up when you need it most (market crashes) and the savings to commissions is for most people not worth the security of knowing you on occasion you are being ‘ripped off’ (legally I suppose) by HFT.

Anon. December 5, 2012 at 6:57 am

How is a Tobin tax going to help provide liquidity during crashes?

Also, the savings don’t come in the form of commissions but tighter spreads and decreased price impact.

Ray Lopez December 5, 2012 at 10:00 am

My thesis is that during crashes liquidity tries up, today, without a Tobin tax. So a Tobin tax will not effect liquidity during crashes, as it is effectively zero. I think the data backs up this view (that during crashes, liquidity vanishes).

Anon. December 5, 2012 at 11:11 am

So what is the point of a Tobin tax if it won’t fix the problematic behavior? Is it just because you don’t like HFTs?

Ray Lopez December 5, 2012 at 12:05 pm

But a Tobin Tax will drive away HFT, at the margin. It will make HFT less profitable, so there’ll be less of it. The assumpion is HFT’s costs exceed the benefits.

Anon. December 5, 2012 at 1:30 pm

How is “less liquidity all the time, including crashes” better than “less liquidity during crashes”?

dead serious December 5, 2012 at 12:11 pm

I have to agree with Anon. You’re not making a very compelling argument for taxing HFT transactions. Also, the Tobin tax appears to be related to the FX market; I’m not sure that covers the lion’s share of HFT activity.

Many people are averse to the notion of levying a”tax;’ instead, introduce (larger) transaction fees. I doubt the markets will seize up. A lack of liquidity hasn’t been a chronic historical problem.

Ray Lopez December 5, 2012 at 2:19 pm

Well I see I chose the right major after all–engineering and not the humanities. So let’s use small words: taxing HFT will reduce HFT. HFT does not add to liquidity during crashes, only non-crashes. We have enough liquidity during non-crashes, and even before HFT HFT costs outweigh any benefits (cost of front running, painting the tape, creating unstable markets due to speculative dark pools, etc)–this is an assumption. Ergo, reducing HFT through a tax–let’s call it Tobin’s tax–is good. Follow?

dead serious December 5, 2012 at 3:01 pm

Use all the small words you want: you’re using the wrong words. Stick with engineering because your logic and writing skills leave a lot to be desired.

Your stated goal seems to be “reducing HFT because it doesn’t provide liquidity.” I didn’t realize that was its primary, or only purpose.

Regardless, I agree with you in principle that HFT doesn’t add any public value and thus those transactions should be subject to the same “rules” as other investors’ transactions (which incur transaction fees).

Anon. December 5, 2012 at 3:03 pm

“We have enough liquidity during non-crashes”

This is blatantly untrue, as evidenced by the gigantic decrease in spreads since HFT algos started making markets.

Ray Lopez December 5, 2012 at 6:02 pm

@anon – “We have enough liquidity during non-crashes” This is blatantly untrue, as evidenced by the gigantic decrease in spreads since HFT algos started making markets

Not true at all that decreases in spreads is evidence of lack of liquidity prior to HFT algos. All HFT has done in non-crashes is increase trading between programs on public exchanges. This did not increase liquidity as evidenced by the rise of dark pools. see: http://en.wikipedia.org/wiki/Dark_pools I am arguing against HFT in public exchanges, not private dark pools.

Mark Thorson December 5, 2012 at 9:33 am

That’s what I was thinking. This is good old-fashioned stock manipulation, given a new name. It needs to be prosecuted, not taxed.

Rahul December 5, 2012 at 10:11 am

What are the sort of activities considered “stock manipulation” in the old-fashioned sense? Other than inside trading what all can an individual pure buyer / seller do that is deemed criminal?

Not being sarcastic; just curious.

Ray Lopez December 5, 2012 at 12:07 pm

Painting the tape: bogus information that distorts the true price-signaling mechanism of the free market. Front running: cutting in line so the real buyer/seller gets a worse price than otherwise.

Mark Thorson December 5, 2012 at 1:35 pm

Traditional stock manipulation consists of making many small trades. For example, if every time the stock ticks up, you sell a little to drive it back down. You can’t overcome a strong movement of a stock this way, but if the stock is just sort of drifting, you can drive it down (or up) by these exploratory trades, then swoop in and make a large purchase (or sale) at the distorted price.

Rahul December 6, 2012 at 12:41 am

Thanks! That just sounds like a strategy to me; I don’t see why the SEC chooses to make that illegal.

Mark Thorson December 6, 2012 at 10:09 am

It’s only illegal if you get caught, and even then you are only forced to disgorge your ill-gotten gains. There’s no penalty and nobody goes to jail.

I suppose these days you could arrange to have a stock manipulation strategy found by some sort of neural network learning algorithm. You could honestly (heh) say you don’t know how the strategy works — the learning algorithm generated it.

Piyo December 5, 2012 at 10:23 am

You may think you can “piece it together”, but you’re wrong.

DayTradingSince1984 December 5, 2012 at 6:46 am
Tomas December 5, 2012 at 10:10 am

A form of exploration-exploitation tradeoff in the unfortunately named “artificial intelligence”.

Sebastian H December 5, 2012 at 11:17 am

The purpose of the Tobin tax isn’t to provide liquidity. Its purpose is to restrict HFT to truly important trades. The main argument against a Tobin tax is that HFT provides liquidity. But apparently HFT doesn’t actually provide much liquidity, so that isn’t much of a defense.

JWatts December 5, 2012 at 11:34 am

“The main argument against a Tobin tax is that HFT provides liquidity”

Unless the tax is necessary and provides some measurable good, then it’s irrelevant why you don’t enact it.

mulp December 5, 2012 at 2:05 pm

When HFT is key to the pump and dump asset inflation that monetizing bad debt until the ponzi scheme collapses and the continued HFT asset churn transfer more wealth to the trading system as the bubble collapses and all liquidity in business and individual money market funds crashes as runs on the shadow banks threaten bystanders with huge losses, and the taxpayers rush in to save their own money from disappearing into the vapor created through the help of HFT, isn’t a tax on HFT justified to recover the taxpayer funds used to prevent all the money of taxpayers vanishing into the pockets of the best HFT traders?

Sure, the government backed institutions starved Wall Street of the mortgages backed by real income and assets, and the HFT wanted far more mortgage backed securities or google share than were on the market, why should anyone meet the demand with bad debt securities at face value and shares of pets.com because the founders of google refuse to put their future in the hands of Wall Street HFT?

Anyone care to predict the date when all the trillions in market cap in the NASDAQ circa 1999-2000 will be restored? When will pets.com return to the hot stock of the day? When will HFT provide the liquidity that pets.com needs to match google’s performance? Actually, when will HFT get facebook up to $250 a share? where is that old pump and dump asset churn to create wealth? Are tax credits for capital gains needed, say match capital gains dollar for dollar with tax credits?

msl December 5, 2012 at 2:46 pm

@Mulp
“When HFT is key to the pump and dump asset inflation that monetizing bad debt until the ponzi scheme collapses and the continued HFT asset churn transfer more wealth to the trading system as the bubble collapses and all liquidity in business and individual money market funds crashes as runs on the shadow banks threaten bystanders with huge losses, and the taxpayers rush in to save their own money from disappearing into the vapor created through the help of HFT, isn’t a tax on HFT justified to recover the taxpayer funds used to prevent all the money of taxpayers vanishing into the pockets of the best HFT traders?”

Are you a human?

mark December 5, 2012 at 3:09 pm

Matt Levine at Dealbreaker had a great follow up on the NYT – reported study, the gist of which is that the study supports a conclusion that HFT costs investors 1 bp.

http://dealbreaker.com/2012/12/study-suggests-high-frequency-trading-might-costs-small-investors-almost-one-one-hundredth-of-one-percent-of-their-investments/

From my own non-scientific perspective, the study Tyler cites fits my experience. Since HFT has grown so prevalent, I actually find it rather easy to pick up a few cents on every trade by putting in limit orders a few cents away from the current market price. Commonly, the bids/offers get taken up rather quickly and I assume it is HFT programs that are making the market move around in such small amounts because of some deep statistical pattern that is lost to the vast majority of us. It is possible they are systematically making money and that the prices I am seeing are in some way distorted. Still, I am not complaining because the limited data I am observing is generally favorable to my interests.

dead serious December 5, 2012 at 4:03 pm

If you’re a retail investor, you’re possibly paying an extra commission for putting in a limit order rather than a market order.

Anon. December 5, 2012 at 4:30 pm

What broker does that? Even as a retail investor I get paid by the exchange for using limit orders (as long as they’re not instantly filled that is).

dead serious December 5, 2012 at 5:26 pm

My apologies. I could have sworn that Fidelity charged more for limit orders. It’s possible that they used to and no longer do. It’s also possible I was completely mistaken.

HFT Trader December 5, 2012 at 4:56 pm

The study cites the fact that the largest and most profitable HFT traders tend to lose money on small orders on average. The paper’s hypothesis is that small orders are used to gain information to generate profit for larger orders. The paper cited that the largest and most profitable HFTs (termed “A-HFTs”) made about $110,000/day trading the S&P futures in their large orders and lost about $7,000/day in small orders (defined as <20 contracts).

I worked at one of those eight "A-HFTs" for several years so I can assure that this hypothesis is not the correct explanation. Rather this phenomenon is because different strategies are silo-ed into separate sub-accounts. An HFT is always experimenting with new strategies and signals. For accounting purposes it will run these in a separate position block with its own PnL and risk limits.

When trying out a new strategy they'll keep the size small while it's untested. Thus the most well-tested and profitable strategies will run with very large risk limits and tend to make very large trades. Then at any given time they'll be dozens of experimental strategies running with small risk limits that will only have the capacity to make small trades. Most of these strategies will prove either un-profitable or will need to be tweaked over time to make money consistently. (By which time once they're tweaked they'll be scaled up).

Thus the small orders from an HFT are going to disproportionately represent those coming from small and generally un-profitable strategies. When one looks at the activity of the market participant in aggregate it will look like it is consistently losing money on small positions as a whole.

eorr December 5, 2012 at 6:25 pm

This makes as much sense if true, he is just explaining that his model explains the same phenomena by explaining the small trades a ms based pump and dump schemes too.

dead serious December 5, 2012 at 7:16 pm

Would you argue that you were providing a public good in the form of liquidity?

Comments on this entry are closed.

Previous post:

Next post: