Tuesday, October 27, 2009

Dark Pools Are Not Scary Shady Places That Rip Off Average Investors

The next step in the discussion about high frequency trading leads logically to the topic of dark pools.  I'm constantly mind-boggled to see the same people who were railing against high frequency trading proceed to dark pools as their next target.  Why is this so surprising to me?  It's simple:  dark pools are the antidote to high frequency trading.  They're where you go to hide from the computer algorithms who are making lightening fast trades in reaction to the bids and offers you post in the marketplace.

Let's take a quick step back.  One component of high frequency trading is what I'll call "pattern mappers" - algorithms that try to deduce how stock prices are going to move based on the bids and offers that are lining up in the marketplace on open books.  For example, if XYZ normally trades one million shares a day, and there is a bid posted for 100,000 shares, it might be reasonable for a high frequency trading algorithm, or any other market participant for that matter, to assume that there is more demand for shares than there is supply, and that the price of XYZ will go higher.  Of course, there is no guarantee that this will happen, and the person buying 100,000 XYZ may not ever change his limit.  So, the algorithm, we'll call her Patty Pattern Mapper, buys stock in XYZ hoping to sell it to the buyer at a higher price.  The big buyer, we'll call him Donnie Dark Pool, has other options though.  Instead of exposing his bid to Patty's pattern mapping algo on the NYSE, he can enter his order in a dark pool, where no one can see the order.  If, and only if, there is someone on the other side of his trade willing to sell shares to Donnie, the stock will trade.

Now, there are some key facts about dark pool trades that lots of people either ignore or fail to understand.   First, all dark pool trades, like any other trades, are required to be reported to the tape within 90 seconds.   Second, all dark pool trades, like any other trades, in accordance with Reg NMS, are required to take place within the inside market - the NBBO - national best bid/offer.  So, no matter how much Karl Denninger would like to construct an example where a stock is trading at $10 on the "open exchange" and there is a seller in a dark pool willing to sell shares at 9.90, which are instantly snatched up by Goldman Sachs  for $9.90 to resell to the open market at $10 - that simply does not happen.  The dark pool either routs the seller's order to the open $10 bid, the stock trades at $10 or better in the dark pool,  or no trade takes place.

Third, and most importantly, trades in a dark pool, or in any other marketplace, only happen when there are matching supply and demand for shares at a given price.  One misconception is that dark pools somehow unfairly allow buyers to buy large blocks of stock without moving the price.  Huh?  Yeah - they can buy large blocks of stock if there is someone willing to sell large blocks of stock.  Otherwise, they can either raise their bid in the dark pool until they find liquidity, or not buy shares.  It's impossible to buy "large volumes of shares" at "small volume prices."

An erroneous critique related to this third point is that "large supply of stock should make prices go lower in an open market."  David Weidner's column on Marketwatch gives an example of this common thought error:

"The problem, of course, is that bulky trades move markets. If I'm at Merrill Lynch and I need to unload 500,000 shares of XYZ, I can place the order in dribs and drabs -- through the multiple public markets out there including the Nasdaq, EDGE and Arca. But that order still is going to pressure XYZ's share price. Also, I'm going to be giving myself away. It also means that XYZ's share price should be lower because there are more shares for sale than buyers. That's how free markets are supposed to work, right?"

Is that how free markets are supposed to work?  If I want to sell stock at $15, large amounts of it, then the stock should trade lower?  No - I don't think that's a requirement of good, free, efficient markets at all.   Although, it is one area of market inefficiencies that the high frequency pattern mappers are experts at implementing.  The truth is that if I want to sell 1mm shares of stock at $15, the price need not go down at all.  The price goes down only when there is no one willing to pay $15 and I lower my limit to $14.95.  Then, when I exhaust the demand at $14.95, the price goes down again when I lower my price to $14.90.  Supply of stock at a given price (say $15) does not make a stock go lower.  Supply of stock at increasingly lower prices without a corresponding match in demand makes a stock go lower - this is a key point, and is not semantic nitpicking.  We're so accustomed to trading off of what we think other traders are going to do that it requires some philosophical forethought to understand this concept.   

In the ideal market, I believe that the entire marketplace would be dark - we need not even see bid and ask prices - just one last price.  Everyone can enter their bids and offers - their supply and demand - into this one big dark pool, and trades would print as matches were made, with no one worrying about being out-traded by Patty Pattern Mapper.

I believe it's totally consistent to have the view that both dark pools and  high frequency trading algorithms which exploit the bids and offers they see on open exchanges are ok.  However, I think it's logically impossible to be against both these pattern mappers and the dark pools which enable other traders to hide from them.  Furthermore, I believe its clear that individual investors are not disadvantaged by dark pools, and elimination of dark pools would result in higher execution costs.



David Merkel said...

Spot on. The big guys have to have someplace where they can bid and offer in size without freaking out the retail market. The trades get reported, and clever traders lean a lot from seeing where block trades go down.

PK said...

I think you are spreading a common misconception.

A lot of HFT algos ping dark pools... They have a pretty good idea of who is hiding in them and they do a good job of modelling the liquidity in them. Dark pools are not the "antidote" they are simply another venue for old-school traders to get nailed.

If a human trader places a large block trade - an algo will either source liquidity away from the dark pool and break up the trade onto the normal exchanges, or it will source cheaper liquidity from exchanges to the dark pool. Either way the algorithm is going to capture a liquidity premium.


Kid Dynamite said...

ah yes... i was waiting for someone to bring this up. Absolutely- HFT algo's are big users of dark pools - for liquidity. but they cannot "pattern map" the dark pools like they can with the visible bids and offers in the "open" exchanges. That's the whole point of dark pools. do dark pools prevent any sort of knowledge of what's happening with the trades in them? of course not - they report all trades, just like other exchanges, which is why people don't need to be upset about "lack of transparency"

the main goal is to avoid the "pattern mapping" algo's that people are very afraid of and angry about - the ones that buy stock when they see other bids, and sell stock when they see other offers.

any way you cut it, taking away the informational edge by eliminating displayed bid/ask sizes is an improvement, even though there is still information to be gained/gamed within these trades

Steve said...

Thanks for another great post.

There was a "pattern map" game played against dark pool traders where small orders would be entered to probe for larger orders. If you get a fill on your 100sh order it is quite reasonable to assume that there's a much bigger order in there to trade ahead of (also reasonable to assume that the contra is simultaneously working this large order in other open and closed venues as well).

At the time, various countermeasures were being discussed. I haven't heard of this practice in a while so maybe it has been thwarted.

While this is "pattern-mapping" it isn't necessarily an HFT strategy because it's simple enough to do manually. (BTW, I'm not digging the "pattern map" phrase. Sounds like you wanted to say pattern *match* but even that may not be the best term).

Kid Dynamite said...

i actually tried to implement a strategy like this about 8 years ago, by assuming that fills i got in the overnight cross were adverse selection - and reversing them... it didn't work.

Steve said...

The biggest problem with the practice of pinging/probing dark pools is that the confused minds of the ignorati (I dig that term!) have morphed this into something that also happens on regular exchanges.

The problem is that it doesn't work because a 100sh fill is *not* indicative of a large order. And then you have 100sh to dispose of.

A couple of months ago Denninger wrote a deeply flawed example of how HFT used pings to find the absolute worst price a trader would accept. The biggest flaw was the assumption that all other market participants went home. He also ignored the fact that his algorithm would also act in cases where there was no size (ie one and done), the algo would accumulate shares on the wrong side of the trade, and it was easily gamed.

The so-called market "professionals" at Zero Hedge didn't notice any of these flaws when they posted a piece celebrating Denninger's analysis. Shows why it's so important (and difficult) to educate people on how these things really work. I don't think the ZH owners have any interest in learning the facts but there's a lot of others who have the capacity to understand and I think you're doing a great job reaching them.

Anonymous said...

You are wrong.

Measuring Arbitrage in Milliseconds