The Micro-Structure of Dark Pools

Session 1: Infrastructure of Digital Economy

3:46 pm – 4:10 pm

Dan Laidley (University of Leicester) – The Micro-Structure of Dark Pools

To watch the video on YouTube, click the ‘YouTube’ button above.

Transcript: –

Dan Ladley

Okay, so I’m going to give a little bit behind. So I’m going to give a very a whistle stop tour of this paper. So this paper was presentations all about market microstructure of dark pools, and particularly looking at the details of how their pricing rules constructed. So essentially the design questions of dark pools. So for those, if you’re sort of less familiar with dark pools, dark pools are alternative trading venues which have grown in size and significance greatly in the last decade. They separate themselves from lit markets and more common financial markets in that they repress information, pre trade, and they do this to a lesser or greater extent. One extreme of the continuum, essentially, they give no information to traders prior to trade. So you can’t see the best you know, any information about the prices or the quantities of orders available in the market, they’re essentially completely opaque. Other markets may give you partial information, they might tell you trade price or without available quantity. But yes, and say they’re dark, they hide liquidity information, there may be various advantages or disadvantages for this in that they can be may allow large traders Well, the argument for them is that the AI large traders to conduct liquidity trades rather than information trades without having an impact on the market as a whole because we can hide that volume. There is important there are of course, no disadvantages with soccer in tune, it’s very hard a priority to separate out liquidity versus information trading. As a result, there are concerns from regulators on the impact of dark pools on information dissemination on price discovery, you know, the fact that trades are happening, or that the orders have been placed that the information of those orders is hidden, and traders can’t see it until trades have occurred, is going to limit the ability of trade of other traders other financial institutions to accurately price and value the asset. Essentially, it’s making markets more opaque than potentially at the cost of other market participants. If these markets were small, that might not be an issue. However, their size and importance across the major financial markets and economies has been growing notably. So somewhere between 40 and 60% of trade occurs through dark pools. So these are major trading venues which have a very large impact on our overall price discovery process. The rate of growth of these trade markets is also very high. So they were around 45% in the US in 2017, but also 50% year on 50% increase over time. The other difficulty we have with dark pools is that there’s still relatively little work for you trying to understand them, they are difficult to approach analytically. And there’s still relatively little information, empirical information available. So our understanding of dark pools is far weaker than our understanding of other financial trading venues. So what I’m going to talk about in this paper, I’m going to talk about whether the existence of dark pools is beneficial to traders and markets, both separately and together. I’m going to talk about where traders prefer should prefer to call to trade should they prefer to trade in traditional markets or whether they should prefer to trade in dark pools. I’m also going to talk about how the design of dark pools matters. There are many design questions for institutions designing and creating dark pools. One of the most pressing is how those how the price is discovered. In a typical divorce a market price is determined by the price of the limit order the order that was in the market. First, the essentially the liquidity giving order. In dark pools, there are many different ways that this can be done. That’s one option. But there are ways to explore other options essentially taking price signals from liy markets we’re talking. There’s alternatives around using average prices, things such as the V whap. So on so I’m going to talk about how the design of dark pools impacts this question.

So I’m not gonna say a great deal about the literature given the time. There are some empirical and theoretical studies looking at this, but as I say the information from it, they’re also very hard to analyse, analytically. So in this paper, we’re going to be taking a numerical approach, we’re essentially going to be setting up a game, essentially trade in the dark pool as a game. And then using a sophisticated numerical technique to identify the mark of perfect equilibria of that game, we’re then going to analyse that equilibria numerically to understand the relevant market statistics. So I’m going to start off setting out this game. It’s we’ve tried to design the situation the model setup to be as realistic as possible within the the bounds of tractability. I should say I emphasise the bounds of tractability. In this case, that whilst we’re taking numerical approach that is doing much more than we’d be able to do theoretically, this situation is still sufficiently complex that there are still limits, though there are constraints on how complicated a model we can solve. So to put this in context, this model takes some aspects of a month on a very high performance machine to arrive at an equilibria. The results we’re going to demonstrate are averaged over contrast that across several 100 runs. So there’s a lot of computational time to try and get these results. So the way the model works, we have two types of traders, traders trading large quantities of the order and traders trading small quantities of the order. So in line with our essentially our the common beliefs around how the different groups of traders particularly used our pools, we have two markets, we have a lit market, traditional, order book lit market, and a dark pool whose design I’ll talk about in a few slides time. I’m not going to say too much about the notation. But the billing market works based on a series of discrete set of prices, traders consuming limit orders to the book which specify a quantity of price and quantity that they wish to either buy or sell, other traders can then submit market orders to execute against those orders. In the dark book, the submission decision is simpler, all they have to do is submit a quantity and whether they want to buy or sell, the price in the dark book is going to be determined by the book, in our case. So it’s going to be determined by the prices in the lit book. So there is some information for the traders switching to dark pool. But importantly, they don’t know what quantity is available there. So if there is if they’re wanting to buy, and there is a seller in the dark book, they’ll execute instantly. If there isn’t, they’ll enter a queue into the dark book. And we’ll have to wait. This waiting is important as it entails picking off risk, but essentially both the cost of waiting and a risk to it. So priority in the dark book, the queue is going to be determined in one of two ways either time, priority or size priority. Both of these rules are used in real financial markets. Time priority is the most common it’s used in lit markets around the world, essentially, whichever order entered the book first, as part of ticket price have priority. Size priority is used in some dark pools. essentially, whoever has the largest order in the book has priority. So this gives priority to large traders over small traders effectively. So I’d just like to sort of give you an idea of the difference information sets available to traders. So in the lit market that given the trader can see the best bid the best ask the depths that those prices, they can deduce the price in the dark pool, because it’s the mid price or the bid in the ask. They can see the total quantity available at the buy side the total quantity available at the sell side. And they’ve got details of previous prices and transactions. In the dark book all they know is the price. And they don’t know for instance, whether the execution would be instantaneous, or whether there’ll be entering a queue. So there’s a very different set of information. The traders submission decision when they enter the market is then which market to which of the two venues to disseminate to what’s the quantity of that border and if it’s a lit market, what the price of it is.

So the traders enter the markets, there’s a, there’s a continuous stream of traders entering the market they enter to randomly determine times with the time between entries being determined by a plus on distribution. Each traders risk neutral. And they arrive with a private valuation of the asset, which determines how they’re going, which is going to affect their submission decisions. They don’t enter, they can’t continuously monitor the market. However, after submitting an order, they re enter again, it’s a randomly chosen point in the future. So the traders decision is both what order should I submit when I enter, but in making that decision, they have to factor in the fact that they will reenter the market. And if they’re not happy with their current order, they can adjust it. So this is essentially it’s a dynamic programming problem. So they use a series of decisions to try and maximise their expected return. small traders buy or sell one unit of the asset Australia’s buy or sell quantity which is greater than one.

The way the, so the traders submit orders based on their strategy, we use a numerical optimization technique to identify the optimal strategy of traders. So they take all the information available at the analyst in the dark book that they can see and submit the optimal order. So skip ahead a few slides. In order to work out the strategy we’re essentially identifying Markov-perfect Bayesian equilibrium. The estimation technique we do we’ve used for this is somewhat lets say on winded and it takes about a month to run. But what we do is essentially optimise the trader strategies such that the traders estimate of the expected payoff from a given order matches their realised payoff over time. So it’s a continuously improving optimization. We know we found an equilibrium when those expectations match realisations. Sorry, I think there’s a there’s a question

Anindya Chakrabarti

quickly, so what is the information set or the agents?

Dan Ladley

Okay, so the information set is all the information about the lit market. So best bid the best offer the five best bid prices five best offer prices, the quantities at those prices and total quantity in the bid side, the total quantity on the Ask side, the most recent transaction and its direction, and the private information about the trader as well. So the public can look like information. And all of that is taking into account and selecting a market price and the quantity. So I’m very quickly going to give some key results. This is easiest with figures. So what we can see here is the distribution of traders so how many traders are trading in the lit market versus a dark market for different levels of volatility in the system. So I’m going to focus on this graph here, which shows it for the markets with both time priority and size priority across different levels of asset volatility. What we can see is that for all levels of asset volatility, the vast majority of traders are in the lit market to appear very few traders are in the dark market. However, as volatility increases, so the market becomes less sure traders migrate towards a dark market. And this is particularly the case where we have the time priority. We’ll talk about a bit more about that later. As volatility continues to increase however, we end up with a migration again back towards the lit market. The reason for this is all to do with traders making a trade off between speed of execution and picking off risk. So under very low volatility, there is the picking off risk is very low. So as a result, traders tend to place orders in the lit market to maximise the speed of transaction and they have relatively little risk from displaying their order in the market. It’s unlikely to be picked off as volatility becomes to increase, it becomes benefits to trading in the dark markets, because the dark market hides information about the traders order, it reduces the picking off risk. What we find though, as the volatility increases even more than migration, we end up with a migration back to the lit market. The reason for this is that from very high levels of picking off risk, the way in which prices are determined in the dark market, so the midpoint of the lit market is not beneficial to traders, it’s better for them to be in the market to place orders either far from the current market price or to take market or the futures market orders essentially benefit from immediacy. So, dark pools are only essentially very popular amongst traders are only preferred by traders for intermediate levels of volatility.

So, the other thing that we need this worth highlighting about dark pools is that they unambiguously increase liquidity costs within the market. So the existence of a dark pool is not beneficial to either the well to the majority of traders in the market or the welfare of the system as a whole. Essentially by fragmenting liquidity, the traders and the system as a whole loses out, actually, the only traders to benefit from the existence of dark pools are what you would call speculative traders. So these are traders who essentially are trading for no private value, they are speculating, trading off the need of other traders to trade. So dark pools as a whole, reduce bother to reduce welfare. However, if we look at the details of the design, the time priority rule is slightly better than the size priority rule in that regard. So besides priority rule is most harmful for try it again, the time priority rule is most harmful for welfare, because he followed by the size priority rule. The reason for that is the size priority rule benefits large traders, as opposed to the speculators. So I think hopefully, yeah, just fitted into the paper into salt. I’m very happy to answer further questions about this. But that’s in a nutshell, those are our findings, dark pools, damaged market welfare for large traders, but size priority rules, do this to a lesser extent than time priority rules. Thank you.

Anindya Chakrabarti

Thanks for the interesting results. There are, I guess many questions? So first Nikiforos then Sheri?

Nikiforos Panourgias

Dan I was just, maybe I just wasn’t sort of paying attention at that point. But when modelling the lake venue, did you take into account say, the market impact? So you know, when there is a large order, you know, the price will obviously adjust? Or, you know, whether it’s a sell or buy? What was that sort of included? And some the second part would be in terms of, say algorithmic trading, and order routing, and so on are the implications for the design of those of that kind of software?

Dan Ladley

Yes, no good question. You’re absolutely right, I skipped over the vast majority of the model and didn’t talk about market impact. It answers your question No, yes. So we fully model the order book in the lit market. So when a large trader enters the market and submits a big order, they take away liquidity from the book, and that has an impact on the market and the price. So essentially, increasing the spread and moving the price. So it fully captures how you’d expect an order book in a dark market to behave and interact, order routing is a really interesting question. We have an extension to this PayPal that we’re working on at the moment that is looking at questions around preferencing and order routing more generally the assumption in here a relatively straightforward around, you know, traders enter and then have the choice between markets, you know, the lit market or the dark market. I think there are very interesting questions. Around intermediation by brokers in this case, and you’re essentially paying for order flow and choosing which order where orders go. So that’s essentially the next paper. Thank you.

Sheri Markose

So Dan, that was very interesting. I mean, the whole business of information being fragmented, you know, centralised exchanges, stock exchange, fighting for their life against such parasitic sort of distilling away from their own centralised platforms. But the point is that maybe you couldn’t cover it in the time, the market price discovery that’s done on the centralised exchanges, you know, it’s not what is done in the dark pools. In other words, they use the prices determine elsewhere, do they not? They’re parasitic in that sense. So they’re there for a purpose, because the statistics showed us you can remind us again, what it is that this is happening in quite a large extent, isn’t it the hiding away of trades into dark pools? So if you can then explain a little more that you said, the nature of the fact that they use the price determined elsewhere in the dark pools, right?

Dan Ladley

I skipped over this, essentially, the detail of this. So one popular and common way to determine the price in a dark pool is to take the midpoint of an equivalent lit market. So trading the lit market directly affects the execution price for the dark pool. What this means is that as trade moves from lit markets to dark pools, there’s less liquidity, there’s less trade in the lit market, that directly impacts the price discovery process and the execution price in the dark pool, you can see that having fewer traders in the lit market means that the price in a real, very real sense will not be as accurate either, whether that is temporarily or due to a larger spread. And that then will have an impact across the whole systems of market so by filling in the dark market, by moving trade to the dark pools with this is essentially the mechanism by which we’re hitting welfare, or by which welfare is hit across the system, because you’re inhibiting price discovery, and you’re fragmenting liquidity as well. So you’re making the cost of trading higher.

Sheri Markose

But can you just sort of tell us, you know, the dark pools are done by private entities, you know, cartels of people. So it’s sort of like piracy, in some sense. Is there any way we could prevent that by regulation?

Dan Ladley

That’s a very good question. Yeah, you’re absolutely right. So the dark pools are often set up by private entities, they, it’s beneficial for them, they charge a trading fee. But they don’t internalise the cost to the system as a whole. This is why there’s a lot of interest in them from a regulatory point of view, because they’re essentially, according to our results extracting welfare from the system, they, you know, I think certainly, historically, some traders felt it was a benefit to them, you know, particularly due to algorithmic trading in the lit markets. Our results would suggest from a system wide point of view that that’s not the case. So we can show that in some circumstances, some traders do prefer to move to the dark pool.