Research on Selective Delay TEC Process


#1

Hi, this is Peter Zeitz, your friendly research fellow at 0x. I thought I would share some research I am doing on TEC processes. While there are a number of TEC designs that interest me, what I am most excited about right now is process called selective delay. I feel that selective delay has the potential to revolutionize crypto trading by delivering dramatically lower prices, more decentralization, more competition, lower trust requirements. Basically, selective delay is awesome and I’m so excited I needed to share right now even though the research doc is in a very preliminary form. The ideas are all there, but it is missing essentials (proper typesetting, a biblio rather than just links, correction of typos, improved style, better organization, etc…) Also it is a bit technical, so it may be difficult to understand. Post questions to help me learn how to preach the selective delay gospel to a broader audience.

Anyways, it is a public link so anyone can post comments. Your comments, questions, and feedback, will help me sustain my current level of enthusiasm through progressive revisions. I plan to write a greatly simplified and condensed version of this technical paper as a medium blog post. Simplifying things for digestion by non-experts is much more difficult than it sounds and your questions help illuminate the best way of doing this.

I will update the link with newer versions as I progress with revisions.

Abstract: Asset markets utilize a wide variety of trade execution pro-
cesses. While the continuous limit order book is by far the most familiar to
traders on crypto exchanges and equity markets, it may not be the best
suited for the development of decentralized exchanges. In this paper, I
evaluate an alternative trade execution process called a selective delay
limit order book. Selective delay is the dominant execution process in the
foreign exchange market and has recently been making inroads among
equity exchanges in the US and Canada as well. The selective delay limit
order book is identical to the continous limit order book, except for one
simple tweak to order processing rules. Under selective delay, the trade
execution coordinator processes requests to create or cancel limit orders in-
stantaneously. However, requests to fill resting limit orders are subject to
a processing delay. This ensures that whenever a race occurs where tak-
ers seek to fill an order that the maker seeks to cancel, the maker always
wins. I explain how the application of this rule leads to a dramatic re-
duction in the equilibrium bid-ask spread, eliminates tendencies towards
concentration of market activity in large exchanges, eliminates risks of
exchange operators covertly trading against users, and facilitates mass
participation of traders in a unified global marketplace.
Preliminary 0x Research Paper on Selective Delay


0x Developer Meeting 2019-02-05
TEC with no order reservations (open orderbook with soft cancels)
#2

I believe the model of higher latency auctions can be valuable in some contexts, but certainly not all. As I understand this is based on the work of the IEX exchange - did they publish a paper on their market-architecture? I see the references to the FX markets [1] . Generally FX market suffer from centralisation because of banks and central banks.

[1] https://link.springer.com/article/10.1007/s11579-018-0218-3

One issue is that there is going to be arbitrage between slower exchanges and faster exchanges. So in some sense the fact that traders want maximum speed is a universal economics law. If you have two prices p1 and p2, and p1 occurred before p1, p2 is going to be accurate by definition (ceteris paribus).I think solutions to issues with HFT trading and microstructure should take into account this fact or law. Say you have a Decentral exchange and central exchange. A market-maker has to take into account the pricing at the central exchange. if he doesn’t he will get arbitraged. MM don’t only quote on a single venue, they have to take into account all liquidity pools. Trading in some fundamental ways is about speed the same way racing is about speed. The more recent price is always more accurate than an old price (nobody would trade at the price of gold 10 years ago and the same principle applies to the price of gold 10 seconds ago). I think the success of IEX exchange is due to other factors - selling a delay as a new technology I find a bit absurd. I generally we would want to have the speed of DEX improve as much as we can to improve competitiveness with CEX while developing the argument more formally in terms of the lower default risk etc.


#3

Relayers could certainly allow makers to post multiple order types, so that a single TEC could support both continuous and selective delay order types simultaneously. That is in fact how IEX works. The selective delay orders are protected from aggressive order flow (arbitrage) and thus support much lower spreads, This protection enables half of trading volume at IEX to obtain execution at the bid-ask midpoint. This ability to trade at a spread of zero is the key user value proposition.

Here is a cool video explaining how midpoint execution and other pegged order types at IEX work:


Here is a document containing details on the IEX process:

We would have a very different architecture that more closely resembles ‘last look’ in forex markets, but the underlying benefit of lower spreads remains the same.

Hope that helps.

Note: Maybe your concerns stem from the volume of arbitrage-related activity. It is true that a lot of activity at 0x relayers is driven by arbitrage directed at mispriced orders available on other exchanges (the AMM in particular lose a lot of money this way). If you are running an arb bot, you could still use selective delay orders to arbitrage mispriced orders at other venues. The low spreads available from selective delay orders facilitate this activity (i.e. it becomes possible to exploit a smaller price differential in an atomic arb op).We hope that selective delay will spur further growth in this activity.


#4

Is there a high level write-up on TEC? I can see the spec here TEC with continuous time matching

I do run arb and MM bots. MM on DEX will use the biggest liquidity pools otherwise he will get arbitraged by other MM. so if MM_1 quotes eg 99/100 and is by default slower then another MM_2, and MM_2 moves his quotes to 101/103, then MM_1’s quotes are stale. I can’t see the utility of a delay in MM code under any condition. “Arbitrage” is not only an activity, it refers to the fact that if a MM provide quotes the MM better be accurate and not introduce possible arbitrage. The arbitragreurs profit is the market-makers loss. The MM can be protect by quoting much wider spreads. He will be outcompeted by other MM who can offer lower spreads. In any case an MM subjecting himself to delays will lose against those who don’t.

IEX is quite specific case which also underlies the very heterognous market structure of US equity markets. I will read more about it.

Maker makers want speed and high throughput. I am very glad 0x introduced the MM program so there is this conversation. Any exchange that wants to be successful has to cater to its users, and MM/HFT are the power users driving volume of an exchange.


#5

To be clear, the MM is not subjected to a delay. Only the taker side is subjected to a delay. This delay on the taker side protects the MM from arbitrage.


#6

I am interested in this line of research - I think the problem space is very interesting. I would be particularly interested in research of “slower” decentral auctions for particular type of assets on-chain, and new methods for price discovery (sealed-bid and similar cryptographic methods). So I can see a lot of utility of “synchronized auctions” for e.g. ICO sales (see Gnosis).

It seems to me there are a few premises in the paper & line of research that don’t hold for continous auctions. I have been reading the MM literature recently and implemented actual algorithms in the market so I’m quite aware of the short-comings. One of them is the assumption of a single liquidity pool. MM have to use as input the general market-conditions from all major liquidity pools.

Maker-makers engage in arbitrage and can be on the taker side as well. So there is not necessarily 2 groups which uniformly are on one-side. When MM posts an order he has to wait which is the main disadvantage over crossing the spread. HFT would cross the spread to gain speed. So maker and taker orders are tools to get to the goal. The fee structure and other features of a market can skew the actions of traders in a certain direction. I have opensourced simple market-making strategies for the public became more aware exactly of the process. Anyways, that’s going a bit off-topic.


#7

Just echoing @0x_peter’s recommendation to consider a selective delay approach akin to IEX’s bid-ask midpoint. Another type of arbitrage impacted by race conditions is spread-chasing when order matching for open orderbooks (like Radar Relay). Currently, the few profitable opportunities are encumbered by this arms race on gas. This increases the minimum of profitable spread values (negatively impacting traders). A TEC without selective delay would only shift these dynamics towards HFT. It might be an overall improvement but these new dynamics might end up putting small arbitrageurs out of the game.