The Cost Of High Frequency Trading

Tyler Durden's picture

Lately, as the topic of High Frequency Trading has gotten front page prominence, there has been much confusion as to the top line impact on traders that utilize HFT methods, and inversely how much of a "toll" on investors high frequency trading is. In other words: what is the cost of liquidity?

Some, such as the TABB group estimate this to be roughly $20 billion a year. Others, such as John Hempton believe this number is a huge exaggeration, and that HFT is practically a philanthropic act of shareholder-focused pro bono activity, practiced on behalf of benevolent banks and magnanimous market makers, armed with massive collocated computers.

Zero Hedge decided to take a look at the facts. Conveniently, none other than specialized brokerage ITG (whose CEO Robert Gasser incidentally recently banned 3rd party algorithms from accessing its POSIT dark pool, and told analysts on the day of
the decision that 'third-party dark aggregation has not been beneficial
to our institutional POSIT constituency.' One wonders why SIGMA X is an exception in this regard) provides a historical trend in what is known as Implementation Shortfall costs, which allows one to determine HFT liquidity-provision costs.

Implementation Shortfall, also known as Slippage, is the toll HFTs collect from investors - this is, on average, the cost of spread and frontrunning. As per ITG's definition:

Implementation Shortfall (IS) Costs – comprised of 2 pieces:

  1. Timing Delay Costs - Any delay cost incurred between the Initial Decision (Open on Day 1) and the Broker Placement Price. Think of this as the cost of Seeking Liquidity.
  2. Market Impact Costs - Price change between the time the Order is placed with the Broker and the eventual trade price.

What is the numeric representation of IS costs? ITG provides a historical breakdown here:

The first immediately obvious thing is the gradual decline in Commission Costs over the past 5 years, as traditional venues for trading have been replaced with commoditized, HFT trading protocols: this has forced agents to compete based on cost, driving commissions ever lower: without doubt a benefit to investors who have reaped this one particular benefit of HFT.

Yet what is troubling is that IS costs, which have been relatively flat over the past 5 years, skyrocketed over the second half of the past year, with a particularly notable jump in Q4 of 2008 - the days of the turbulent and volatile equity market post the Lehman collapse. Also, the dramatic increase in IS, which averaged 53 bps in 2008, is among the primary reasons why HFT operators in 2008 had a record year, and why firms like RenTec had a massive return difference between its market-neutral HFT strategy (Medallion, with 42% annualized returns since inception and a 5.3 Sharpe ratio over the last three years) and its long-short 175/75 quantitative strategy (RIEF, with deplorable returns). Also, the recent ramp up in HFT by firms such as Goldman Sachs via the NYSE's Supplemental Liquidity Provider program may be a major clue as to the record number of $100MM + trading days in Q1 and soon, in Q2.

So what is the bottom line? Observing the volume of NYSE stocks traded globally one gets an average number of just over 6 billion share daily.

Taking the 6 billion daily average, and applying an average stock price of $20/share and using the assumption that HFTs represent 70% of the volume, while capturing 50% (and according to Zero Hedge sources, this number could be as high as 90%) of the 53 bps average IS spread in 2008 results in HFT slippage capture of over $220 million daily on NYSE stocks alone. And this does not even count rebates. Taking this one step further, assuming 250 trading days in a given year, brings the total annual costs to investors (and revenues to HFT strategies) to over $55 billion. If one expands this methodology to non-NYSE member stocks and exchanges, a reasonable guess for HFT tolls in the US alone to be over $100 billion.

One can see why HFT is a sacred cash cow for the limited group of participants who benefit from "providing liquidity."

Recently, the market cap of U.S. listed stocks was just under $12 trillion: if the total annual cost to investors is indeed in the $100 billion ballpark, which they fork over to the likes of Goldman, RenTec and others for providing liquidity, it means that the liquidity premium of stock trading has increased to 0.8% of total assets.

Of course, arguments can be made that this is a fair price to pay in exchange for having daily liquidity  available (although the debate of whether liquidity=volume is highly relevant) at our fingertips provided by thousands of computers which trade millions of shares every second. The bigger question is what is the potential for abuse of this highly under the cover P&L item, and have firms such as Goldman simply become a toll aggregator as a result of persistent stock churning and liquidity provisioning. As $100 billion is over 0.5% of the recently released GDP, does it not make sense to more actively evaluate whether the HFT market would benefit from some substantial anti-trust intervention, which in turn would promote higher competition and the break up of the highly profitable trading monopoly of a very select few?

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Anonymous's picture

You are missing one of the major sources of slippage: the arms race. Traditional asset managers like mutual funds and pension funds are spending hundreds of millions probably billions on computers, consultants, and software to keep up with the rapidly evolving trading process.

RobotTrader's picture

High Frequency Trading will come to an abrupt halt when too many of these gamers micro-trading "leading stocks" along a momentum line get shanked when the stock is slammed on an earnings warning.

Imagine some 22-year old short-term trader gunning his computers to buy and sell a stock thousands of times a day, earning vast profits, only to see 3 months of gains wiped out in one fell swoop when the stock gaps down on earnings (e.g. AMZN, AKAM, SYNA, FSLR, etc.)

Anonymous's picture

Yeh, too bad it doesn't show Fedor coming back 30 seconds later and putting Kevin in a kimora (if my memory serves me correctly, don't know how to spell it either).

Miles Kendig's picture

And the hits just keep on rollin' along for the chum churners...

BTW, excellent as always..

Anonymous's picture

Robotrader, much more like they hit YOUR bid 1 ms after the headline when their keyworded newsfeed printed the bad earnings.

So it will actually be YOU who gets blown away.

sorry, you're such a nice bot.

Anonymous's picture

these programs do not run overnight... they are SHORT TERM...

BorisTheBlade's picture

Emelianenko vs Randleman, love that fight.

Anonymous's picture

Clearly, because you're forced to trade every stock every day regardless of any major events that everybody's aware will take place.

Dr Hackenbush's picture

It's RIGGED... when a stock shows 2 million shares traded, and you figure that is enough volume to be a non-manipulated stock, think again!  In actuality there are probably only a few thousand shares being traded by actual humans. 

Institutions (gansters) simply divide up territory... technology changes nothing but the avenues they roam.

Anonymous's picture

Floor specialists need not apply.

Anonymous's picture

Notice that until the last year's market hiccups there's been a long-term downtrend in these costs. This is the positive effect of HFT. This graph only shows the tail end of a decade-long HFT-led reduction in costs. In the old days Saluzzi and his gang would take 12-25 cents per share - now the HFT players compete over fractions of a penny per share. It's not philanthropy but it still benefits retail and institutional clients greatly.

Dr Hackenbush's picture

Um, Have you ever watched these bots in action?   They are NOT into price discovery, rather they tend to move the price away from bid/ask pressure.

How can that be of benifit to any TRUE investor?

RobotTrader's picture

Friday's top gamed stocks by dollars traded:

My prediction is that we will soon see SPY trade over 100 billion dollars daily.  That's a lot of lotto tickets flying around.

                    SPY  SPDR S&P 500 ETF   98.81   +0.14   +0.14%   207,357,955   20,489,039,534   BAC  Bank Of America Corporation   14.79   +0.82   +5.87%   375,051,946   5,547,018,281   EEM  iShares MSCI Emerging Markets Index ETF   35.78   +0.17   +0.48%   119,533,753   4,276,917,682   C  Citigroup Inc   3.17   +0.03   +0.96%   1,107,978,233   3,512,290,999   IWM  iShares Russell 2000 Index Fund   55.57   -0.23   -0.41%   40,482,436   2,249,608,969   XOM  Exxon Mobil Corp   70.39   -0.33   -0.47%   28,071,609   1,975,960,558   SDS  ProShares UltraShort S&P500 ETF   47.261   -0.149   -0.31%   38,352,247   1,812,565,545   FAS  Direxion Shs Etf Tr   57.57   +1.006   +1.78%   31,323,593   1,803,299,249   XLF  Financial Select Sector SPDR ETF   13.01   +0.13   +1.01%   131,306,212   1,708,293,818   JPM  JPMorgan Chase & Co   38.65   +0.18   +0.47%   39,089,490   1,510,808,789       
Anonymous's picture

You estimate that HFT represents 50% of the shortfall but this is impossible to estimate.

Consider: An investor makes a decision to buy a stock because he thinks the price will rise. If it rises before he buys it that is shortfall/slippage. But if it rises it means he was right! Maybe most of the shortfall was due to superior stock-picking ability.

Anonymous's picture

Front page of FT has a video of Joe Saluzzi.

Joe is great.

buzzsaw99's picture

Tapeworms provide volume to the digestive system, do you want one?

Anonymous's picture

I'm going to take everything out of the Stock market and other venues that are sacrificed on the altar of inside information and investing it in true market demand and supply driven ones in cocaine, marijuanan, baseball cards, comic books, and tulip bulbs.

Maybe spices,silk,tea and bananas, too.

This electronic trading thing was doomed from the start to become as artificial as Loni Andersen's tits.

Anonymous's picture

Loni Anderson's tits are fake?

Anonymous's picture

as fake as Burt's hair.

Vince Vance's picture

You picked the worst example imaginable. Loni Anderson once had breast reduction surgery.

Anonymous's picture

As fake as "RobotTrader's" avatar above.

Anonymous's picture

One wonders why SIGMA X is an exception in this regard

Their program that gave them the edge walked out the front door in a briefcase. Now it is out there and once broken down, will be used to compete with them. So limiting access to the data is the next logical step to continue to keep your edge.

Miles Kendig's picture

Ya really think so..

Anonymous's picture

Your projections are based on a firm that doesn't allow access to its dark pool.

I would expect their slippage number to increase dramatically as a result. Does the increase coincide with their policy change?

Extrapolating from this source is dramatic, but misleading.

If ITGs 53 bp is the industry norm, then you may have a point. If not this analysis is garbage.

peterpeter's picture

Tyler, I've got a number of problems with your back of the envelope.

1) the assumption that HFTs represent 70% of the volume - is based on the TABB report.  I believe that this number is inflated, but will have to leave it at that, as I have no data to justify that position.  What should not be in question though is that as the overall volume changes (denominator), the percentage attributed to HFTs will also change.  To make the point - if no humans traded for a day and volume dropped to 1M shares, it would be easy to assert that 100% of the trades on that one particular day were from computers, trading with each other (presumably 500K adding liquidity and 500K removing liquidity).  The 100% of course is meaningless in that context.  I bring this up because in Q4 2008, there were many humans making trading decisions, and I would hazard a guess that a much lower overall percentage of trades were from computers than are say in Q2 2009 where we are seeing relatively low volumes.

The point I'm trying to make here is that you can't use the 70% figure (even if you believe that it is correct, which I do not) as a static figure in multiplying it against the spread figures.  The 73% from TABB using whatever method they used was likely a peak figure, and the 53bps is a peak, but the 2 likely happened at different times.

2) I don't know how the IS figures are calculated... and when a trader decides to make a trade and what price they get is of course a function of volatility and the latency on the trade being placed.  If these orders were phoned in, then the increase in IS could just be due to higher market volatility in the face of slow order execution.  I will attempt to churn through a series of bid/ask spreads for the entire US equity market and come up with a data series later - as this article has gotten me interested in what's happened to the spreads, but I don't think there will be material differences in posted bid/ask spreads on US equities over the last year.  Of course, it may make a difference as to how each person doing this analysis weights the various components (volume transacted, market cap, even weight).

3) "And this does not even count rebates."  Nor does it count for the SEC, clearing fees, fees paid to the ECNs for data and connectivity... all of which help account for the reduction in commissions as seen by the average retail trader.  The rebates are greater than these other fees, but not by a whole lot.

4) "Taking this one step further, assuming 250 trading days in a given year, brings the total annual costs to investors (and revenues to HFT strategies) to over $55 billion".  Now, lets say we're with you to this point in time and ignoring all of the fuzzy numbers - what would the costs on spread have been had the computers been shut off?  The spreads would have been wider, but no one can say how much wider... and that is a benefit to the retail investor which you do not mention nor attempt to quantify.

Tyler Durden's picture

w/r/t point 4, i agree and you are right - my point is that this liquidity provisioning is being focused in the hands of a few who could potentially find incremental loopholes to increase IS even higher as the oligopoly shrinks (the latest campaign against FLASHing could be indicative of what to expect by NYSE, NASDAQ, etc - not saying it is right or wrong) which is very concerning. But yes, there is a tradeoff which is why the post is subtitled "the cost of liquidity."

Anonymous's picture


I haven't had a chance to sit down and respond to many points of this post but I will pose one question for now. What was the median order size of this study that experienced 53 bps of implementation shortfall slippage? What was the median ADV of the names in the study (statistics can be manipulated to convey any message you want.. a 100k share order slipping 53 bps in GE is terrible, a 100K share order slipping 53 bps in Google is pretty damn good).

What do you think the median RETAIL order size is? Do you think that order size experiences 53 bps of IS slippage? I understand that the IS slippage of institutional (mutual fund) traders is a direct cost to the retail investing public, but I think at the median, its the cheapest its ever been.

Aside: In the agency execution business 5 Years ago slipping 2 cents off of VWAP was considered success, now VWAP is a commodity. VWAP is an obtainable execution benchmark for large orders, IS is NOT.

Anonymous's picture



Anonymous's picture

Too subtle, Your key point is valid and should be included in the headline in each of your analytical posts.

this liquidity provisioning is being focused in the hands of a few who could potentially find incremental loopholes to increase IS even higher as the oligopoly shrinks

This is the alarming theme underlying all your coverage, and bears constant repeating. Your expanding fan base is still climbing the learning curve here.

Anonymous's picture

the NYSE policies of implementing volume increasing policies that are meant to 'stabilize' the market. prior to the market these policies were not as outright solicitious, but it's not like HFT didnt exist prior to the crash. And it's not like anyone is stupid enough to think that HFT can be 'regulated' out of existence.

An event like this crash had the potential to actually destroy the NYSE and drive trading permanently to other platforms. the NYSE responded by trying to make it look like the recession is not so bad and that 'everyone' is still in the market with its SLP -volume rebate program to get market insiders to increase their trading volume. so the NYSE saved its own ass. that's all. 'rebates' don't make the NYSE any wealthier, they pay off goldman and others to retard a decreasing revenue situation and possible destruction of the NYSE. you really can't blame the nyse. the stock market is not a public entity. and while regulated, it still needs to fund its own existence.

if you want ot get rid of colocation to prevent front running. then go for it. it seems like a smaller tighter argument to say that front running is illegal, and on the principle that the appearance of justice is important, the NYSE should be prohibited from allowing any computers to be located on the exchange premises for a fee. the colocation scheme must be open for free to anyone, or must be closed.

. . .'s picture

anon 22395:  And it's not like anyone is stupid enough to think that HFT can be 'regulated' out of existence.


Technically, it is easy to regulate high frequency trading out of existence.  You just impose a securities transaction tax on the gross revenue that is greater than the spread someone makes from the transaction.  Longer term trades shoot for higher spreads, so they wouldn't be affected much.  It would be like shutting down spam with a tax on email.

One can debate whether shutting down HFT is desirable or politically possible or likely.  But technically, shutting it down is not rocket science.

Anonymous's picture

Why don't E-Trade and the like pony up for co-located servers? Not a huge deal if their commission rate goes up by a couple of nickels. The big boys have to pay up for that service. Why do you expect to get it for free?

Anonymous's picture

ECRI uses government numbers. So..the recession is over.

Milton's picture

BATS seems to be flagrantly violating the rules about locking the market. They give an example (para #3) of going high bid which matches the offer; but, what about going low offer that matches the bid? Wouldn't that violate the uptick rule?

Also, what's with Joe Ratterman denying "UNIQUE" front running?

Anonymous's picture

BATS can make it's own rules. It's a seperate exchange.

zeropointfield's picture
zeropointfield (not verified) Aug 2, 2009 2:08 PM

From Reuters:

But Goldman is talking only about high-frequency trading of stocks, not options and commodities. In options trading alone, Goldman’s algorithmic-driven platform is estimated by a market source to account for 15 percent of the daily trading volume.

By comparison, the high-frequency options volume leader is the giant hedge fund Citadel Investment Group, controlling some 25 percent of the daily trading activity.

Additionally, there’s a great deal of latitude for firms to decide what it considers to be proceeds from high-frequency trading. Bernard Donefer, a professor at CUNY Baruch College and a critic of automated trading strategies, says “nowhere in the market is a trade marked as a high frequency trade.”

. . .'s picture

Zirp:  But Goldman is talking only about high-frequency trading of stocks, not options and commodities.


There's also high frequency trading in FX, rates, and bonds, which Goldman isn't talking about in claiming that less than 1% of its revenue is attributable to HFT.  And by pure coincidence -- no doubt -- FX, rates, and bonds are done in GS' big money maker (lately), the Fixed Income, Currency and Commodities Division.

zeropointfield's picture
zeropointfield (not verified) . . . Aug 2, 2009 3:15 PM

right. so they are running a multidimensional strategy. some of it is linked to the equity (shares) market. It is the underlying for many things, so they you can influence where it goes, they can make a lot of money in the other (e.g. options) markets.

in short, we are not analyzing the market activity of people, we are analyzing the algorithm (and the thinking of the programmer(s) who wrote it).


vicelord's picture


Anonymous's picture

Right on point, Tyler. Thank you!

Anonymous's picture

John is just trying to get publicity. The only problem with it is that he is wrong and will stand by being wrong even in face of things like facts. You would think someone from a "fundamental background" would appreciate fact over opinion which he frequently falls back on. Ego's can be an ugly thing, I would hate to be an investor in his fund they usually end up blowing up because they don't concern themselves with facts but rather opinions.

Ando's picture

The increase in implementation shortfall in the last year may just be due large investors all trying to buy the same stock at the same time or sell at the same time like last fall. I don't think its because HFT have gotten any better at getting profits. Also, many of these recent articles on HFT seem to imply that most of the profits from HFT is going to a select few but I think mutual fund and hedge funds also are large participants in high frequency trading or liquidity providing.
Also I might add the obvious: implementation shortfall for retail stock pickers is close to zero. The cost of implentatuon shortfall is going to large institutional participants who want to make a large transaction in a single day. 50bps is a pretty reasonable cost for the ability to get that kind of liquidity anytime they want it.

SWRichmond's picture

I was able to catch the last half of this interview, thanks a for having him on.  I appreciate his clear explanations of the issues.

e1even1's picture

anyone who uses market orders or stop orders is naive. they shouldn't be trading until they wise up. using a limit order to match a bid/ask, depending on which way you're going, results in the minimum slippage.

slippage is not some new bogeyman. the bid ask spread is a fact of trading life. it always has been. this controversy is just high frequency bellyaching. some people swim with the sharks. others like Mr Saluzzi are over squawking with the chickens.

Tyler Durden's picture

ITG data comes from Plexus subsidiary (a supplier of transaction cost analysis (TCA), trade research and consulting services) which captures 50% of traditional buy side data.

Anonymous's picture

Thanks, so the 53 is based on a data from 50% of the buy side.
I'm not a statistician so I don't know if that sample size is adequate to support your conclusions, but thanks for the disclosure.