Father of HFT: Slow Down

Thomas Peterffy is one of the pioneers of automated stock trading. From NPR’s Planet Money:

…The company he started, Interactive Brokers, does electronic trades on a mind-boggling scale. Forbes estimates his net worth is now over $5 billion.

Peterffy says automation has done some very good things for the world. It’s made buying and selling stocks much much cheaper for everyone.

But Peterffy thinks the race for speed is doing more harm than good now. “We are competing at milliseconds,” he says. “And whether you can shave three milliseconds of an order, has absolutely no social value.”

Here are previous MR posts on high frequency trading.


But him shaving off 10 milliseconds provided social value to the north of 5bn USD...

There's only so much room in Greenwich, CT.

What's his actual argument though? Are all improvements in trading speed wasteful? Probably not. At worst it seems like a small loss.

I'd bet it's the business now requires larger capital investments than he can afford or generates large enough returns to cause the big boys to want to play in his sandbox.

But the small loss to the economy overall can be a large loss to a specific player in the market, say Peterffy's company Interactive Brokers. Higher frequency trading probably increases the risk to his company if the programs don't work as expected so that they make a lot of bad trades before one can pull the plug.

I'd say the marginal benefit of speed improvements declines sharply at some point, and could easily go to zero.

Going from three months to a few minutes helps a lot.

But what does going from one second to ten milliseconds do for us? Isn't the fundmental idea that equity markets encourage investment by providing liquidity to investors? At what speed is there nothing more to gain along that line?

Since when do we ask what an economic activity freely undertaken "does for us?"

Is there are social harm? If not, who cares?

There is social harm in the form of extreme opportunity cost: All the effort of brilliantly smart people designing and using HFT systems could go towards a productive endeavor, or at the very least towards leisure time for these people.

Like digging holes and filling them in, HFT creates work - but the work ultimately has no economic utility apart from random zero-sum monetary transactions.

The comment was broadcast in the context of the "Flash Crash." I suppose the argument is that the speed arms race has increased risk in the stock market: allowing for arbitrarily fast trades is a high risk, low reward venture. Don't know if I believe it, but I think that was the point.

Is speed the problem or automation that is insufficiently controlled?

Opportunity costs? Extremely brilliant people who could be doing socially useful things are instead shaving milliseconds off of trading speed.

On that basis we should ban the olympics too. After all, what social benefit comes from someone running 0.01 seconds faster?

Odd that, given the two facts below, there isn't a bit of skepticism about
Mr Peterffy's motives. Could it be that he doesn't like competition? Just saying ...

1. Today, Peterffy is one of the big players. The company he started, Interactive Brokers, does electronic trades on a mind-boggling scale. Forbes estimates his net worth is now over $5 billion.

2. Peterffy is in favor of more regulation. He'd like to see rules that slow things down.

He's obtained a significant advantage and now he'd like the government to lock in his advantage.

It's more about his method to game the system only works when he's the only one gaming the system.

HFT is all about riding stock momentum, taking a little bit off of the potential profits of knowledgeable traders making intelligent trades. But too many HFT have swamped the system, so the knowledgeable traders have all left for greener pastures.

It seems to me that the problem is less about speed and more about algorithmic trading. Even if the trading happened somewhat more slowly, a poorly programmed trading strategy could still do a lot of harm rather quickly.

Harm to who?

If an algo misprices an xbox to $10, and you buy it, are you harmed? If it misprices to $10k, and you buy it from someone else, are you harmed?

Seems like harm is properly internalized to me.

How much less efficient would the market be if you set a speed limit on the hft at .001 seconds and how much could that reduce the risk of a flash crash? I'd really like to know.
Government regulation could stop the strange consequences caused by companies trying to go faster.

Starting at 13:20 http://www.ted.com/talks/kevin_slavin_how_algorithms_shape_our_world.html
I can't imagine how building these is for the social good.

I think the proper question is "what is the cost to society for the flash crash"? After all, it took the market all of seven (7!) minutes to correct the flash crash. Legislators have discussed what to do about it for two years. What is the cost of that?

I think that the flash crash is the perfect example of how free markets work and how they work so much better than regulators.

Sure, there could be some simple rules that would decrease the probability of flash crashes, such as not allowing entering of orders more than x% away from the market price. In markets where these rules are applied, the flash crash didn't happen or happened to a much smaller degree. But let each market set their own rules for that. The idea that regulators would know which rules wil work best in the future is ignorant.

Let's hypothesize that the answers to both are "little to none," which seems likely. What do you propose?

100 milliseconds would have a significant impact on liquidity and trading costs. I know this sounds very counter intuitive but bear with me.

One big reason that the spreads have declined so much and liquidity increased so much is because market makers have become much much better at managing adverse selection. That is to say separating uninformed flow (like Grandma liquidating some of her 401k to pay this month's expenses) from informed flow (a sophisticated hedge fund with inside information). The higher the ratio of informed, or adverse, flow to uninformed flow the wider a spread a market maker will demand to compensate for his costs.

At a basic level equities can be broken down into two sources of volatility: market beta and idiosyncratic volatility. I.e. a stock may move in line with the broader market or move for company specific reasons. (Of course we can extend this to multi-factor beta models easily).

Gauging company specific movement is difficult for a black box trader. But gauging market beta is easy, simply look at a more liquid security that tracks the market index, e.g. SPY or S&P E-minis. Since these instruments are vastly more liquid than even the largest cap names their price movement tends to lead the movement in corporate single names. I.e. if a news announcement comes out that drives S&P up 50 basis points in 0.5 seconds, a stock with a beta of 2 may move up 100 basis points but take 2 seconds to do so.

The logic behind this simple fact is the driving force behind what has been some enormously profitable strategies in hedge funds and prop desks. Statistical arbitrage and index arbitrage are the two classical examples here. Adverse flow in the form of market beta has historically represented a major cost to market makers.

The advent of very fast computers though gives the market makers an advantage they didn't have before. In effect they can calculate a real-time fair value for many securities based of recent price movements of more liquid securities. So if you're market making MSFT and S&P ticks up, you can modify your quotes at new prices.

This real-time adjustment of quote prices is the primary reason why cancellation rates have risen so quickly. It has very little to do with "phantom liquidity" or market manipulation or anything nefarious (if anything the presence of HFT has made the risks to detection and hence exploitation of an attempted manipulator higher). Rather it has to do with market makers and liquidity providers using real-time information from more liquid instruments to continually update and adjust their quotes. From a black box perspective it looks like a mass of quotes are getting entered and cancelled almost immediately. In reality the vast majority represent a single "meta-order" that might be adjusted dozens or even hundreds of times before being filled.

So what does this have to do with the cost of trading and the impact on liquidity? A minimum time threshold forces market makers to keep quotes out of sync with the broader market. Essentially stale quotes drastically increase their adverse flow ratio. To compensate spreads will widen and liquidity will dry up. You're transferring money from market makers and ordinary traders to a small class of HFT liquidity takers who specialize in picking off stale quotes. Rest assured the vast majority of favorable stale quotes will go to these specialized traders, meanwhile ordinary traders will get the vast majority of unfavorable stale quotes.

Well, does 100 milliseconds really make that much of a difference? I've studied market micro structure at the most intimate levels. During periods of major news, or even simply during the open and close, the market can easily move ten ticks in 100 milliseconds. These periods are also where you have the higher ratio of adverse flow, so market makers (and consequently ordinary investors) will be hit especially hard.

HFT Trader thank you for the thorough answer.

The purpose of the stock market is to raise capital for investment (or allow early investors to cash out). The HFT algos aren't helping investors execute their trades faster, they're front-running them or just trading between algos. Innovaters like IB reduced trade times and costs. I like that I can fill an order quickly when I trade, but there's zero advantage to me from computers raising my trading cost by filling the market with fake orders until it hits one.

" there’s zero advantage to me from computers raising my trading cost by filling the market with fake orders until it hits one."

How are they raising your costs? They are lowering my costs by decreasing the spread significantly. That is very valuable. It's also, as you noted, an advantage that an order is filled more quickly, even though I personally see the value of that is very limited. If the time it takes to fill an order drops from one second to one milisecond, I make no extra money. If the spread decreases from 0.5% to 0.1%, I make on average 0.4% per roundtrip. That is real money that the HFT are "giving" me.

Something I've never quite understood. If HFT offers no value to non-HFT traders, but instead just makes money off them, why aren't some exchanges offering slow trading - eg by executing orders once every x seconds (where x could be 0.1, 1, or 100, or whatever)?

@Tracy I think the exchanges like HFT (more transaction volume) and no individual is obviously harmed by it. The risk is systemic, so no single agent has an incentive to mitigate it.

It seems like it would be easy to solve the problem by imposing a small (possibly stochastic) delay on the execution of each trade. But then the riskiest trades might simply move to other venues.

A tiny frictional tax would probably have the same effect, but traders would REALLY hate that.

The market making algos, afaik, skim tiny amounts off large trades with longer time horizons. In a different day and age, these would have executed with floor brokers and specialists, with opportunity for them to front run these (and for larger amounts of losses). How many people here remember the series of scandals in the nineties where exactly this happened on many stocks.

This seems like established player trying to hobble competition.

For what it is worth, I have an IB account. They are a really good brokerage to try somewhat automated stuff with cheap retail pricing. But they do subsample (snapshots, not ticks) their data feed, in the interest of being able to keep up in periods of high tick volume. It is a useful niche, but should this be the only one?

What is "social value" and how do we measure it?

To the extent that it means something intrinsically moral, aren't there lots of businesses with no social value?

I see lots of potential issues with HFT, but they are mostly around the safeguards to make sure nothing goes amuck. But I don't see any inherent risk in trading faster.

This post makes a simple mistake that's made time and time again. Electronic trading is NOT the same as high-frequency trading. This guy may have been a pioneer of the former, but he has little to nothing to do with the latter. Nor does Interactive Brokers.

Also it's erroneous to believe there was one moment in time where HFT was created. It's present form represented a long-term transition of statistical arbitrage strategies to shorter and shorter time periods. If credit must be given for the "father" the candidate would have to be Renaissance Technologies.

They built the earliest things that resemble modern HFT in the late 90s from their already successful stat arb business. The spread and popularization outside RenTech was mostly due to a network or early pioneers that circulated around the Chicago prop-shop scene. Prominent firms here would include Getco, Jump, Citadel, Hudson River, and Teza.

Long story short Interactive Brokers has a vested interest in suppressing HFT. Their client execution arm is running into trouble against more sophisticated execution services from HFT firms that use better algorithms. (Getco chief among them). These execution services can pay more to retail brokerages for client order flow and IB doesn't like the competition.

I'm pretty disappointed that Alex, who normally has a very good BS detector when it comes to these types of things, was hoodwinked by the fluff piece that NPR did.

Settle down, HFT Trader. Alex just linked to the story. This is how blogging works. He made absolutely zero comment on it. I don't see how he was "hoodwinked".

Comments for this post are closed