Forums  > General  > Microstructure effects for a market maker?  
     
Page 1 of 1
Display using:  

HockeyPlayer


Total Posts: 125
Joined: Nov 2005
 
Posted: 2014-01-29 18:19
When building a market maker that rests limits orders on both sides of the book, what microstructure effects should we be looking at to price those orders? I’m targeting liquid futures markets, and expecting to trade 1-3% of the daily volume.

Current effects that we find useful include the book imbalance (the ratio of volume on the best bid vs the volume on the best offer), and recent trades (someone removing liquidity on the bid is bearish).

What else should we be exploring?

I don't expect to hear anyone secret sauce, but I thought it might be possible to discuss some widely known effects, such as the ones I mentioned. For example, does the price moving away from VWAP have a microstructure effect?

WayconKidd


Total Posts: 108
Joined: Mar 2010
 
Posted: 2014-01-30 18:27
Let's say there exists a widely-known (i.e. Señor Google can help you find out about it) effect that you haven't enumerated. Why would I want to tell you what it is? This of course is the optimal situation for you, because your recourse is a search engine query.

If the situation is any more dire, you're going to have to do some actual work.

Aleph


Total Posts: 78
Joined: Jul 2006
 
Posted: 2014-01-30 23:30
Why are you looking to trade 1-3% and who is 'we'?

I don't expect an answer to the second question. But, the reason I ask is this: having goals like this imply a certain savvy, or desire to posses said savvy. However, HFT has recently fallen out of favour with the press - so it's probably not because you've read something in the FT. Which leaves me confused as you're presenting as someone who wants to be in a game in which microstructure effects are key. On the minus side, you don't seem to posses knowledge correlated with experience which also delivers this savvy. On the plus side you've come up with what's usually my first recent grad interview question.

So, seriously, what's the deal? More information means I can (possibly) help you more.

TL;DR: The important ones.

P.S. What does 'price moving away from VWAP' mean? What VWAP? When? What's a 'price'? What do you mean by 'microstructure effect'? What will you do with that 'effect'?

P.P.S. I'm not being a dick, I just threw you a bone. However, that's probably all my handlers will allow before they put me back in the cage and pull out the tasers.

rukarin


Total Posts: 52
Joined: Nov 2013
 
Posted: 2014-01-31 01:34
`

radikal


Total Posts: 264
Joined: Dec 2012
 
Posted: 2014-01-31 06:43
There's really not enough information here to say much; does the product trade on other exchanges? Is it even FIFO? What's the relative value of Q position? How predictive is weighted mid?

Start with perhaps the last, how predictive is weighted mid? Depending on product specifics, this is not always trivial to implement, and may require some semi-pita cleaning, but in general:
- Bin Every Market Data Tick into a Relative Bid/Ask Strength Bin
- For each piece of MD, label the next tick
- Solve (any crappy simple optimizer should work) to find an X such that for all your ticks when you add X to bid and ask the next tick predictions degenerate to a fair toss
- Wait, what products are we talking about?

There are no surprising facts, only models that are surprised by facts

HockeyPlayer


Total Posts: 125
Joined: Nov 2005
 
Posted: 2014-01-31 17:14

Let's say we are trading CME crude. So it is a FIFO market that dominates the other places where this product trades. CL trades ~200k contracts/day, much of that as 1-lots.

I put 1-3% in to make it clear this is an active market making strategy that is trying to make small profits on lots of volume.

Regarding VWAP and price, let me rephrase that: When building such a market maker, is it worth exploring the effect of weighted midpoint deviating from VWAP? VWAP can be defined as the last 100 contracts, or everything in the last 5 minutes, or maybe separated into aggressive buys and aggressive sells. I don't know, haven't explored that yet.




radikal


Total Posts: 264
Joined: Dec 2012
 
Posted: 2014-01-31 17:40
Often models do include that information, yeah, for some algos, I might include multiple 'versions':
- vwap of last N contracts, vwap of last 2N, vwap of last 10N etc (vs current wmid or bid/ask whatever)
- ratio of aggressive buys/sells is one of those things that people claim to use and there are abundant papers going after edge using this info, but I've never found it terribly educational in any market I trade

I think it's important to consider what your edge is re this trade. It sounds like you're building a tick-taker type strategy with some loose hand wavey mean reversion assumptions...which is totally fine, but for you to do this successfully, you're going to often be long from the offer or short from the bid and instead trade your FV band around the current tickwide market. (This is attempting to "scalp the nosie" generally and, despite how clever you are, is, well, hard. I'd argue that a lot of the firms who succeed at this are getting their edge in many other tangential ways doing this.)

If this is not what you want to do, and you instead want to constantly be on every bid and offer vying for opportunistic fills by queue position, you should realize THAT game is a bit different, and the relative value of having a lot of "smart" microstructure features is low relative to understanding tricks of the exchange and general latency.

Obviously CL has significant low latency competition due to the various number of available arbs (ICE) and correct me if I'm wrong, but I thought that these implied match vs the other months, so there may be significant work in optimizing around safe-Q issues related to implied matching.

ps I trade secrets for alcohol

There are no surprising facts, only models that are surprised by facts

signalseeker


Total Posts: 237
Joined: Oct 2006
 
Posted: 2014-02-01 01:20
I completely agree with radikal here. Depending on what you are actually trying to do, you may want to account for cme and latency related idiosyncrasies before focusing on predictive features.

On the latter front, of course there is a lot to explore and someone is very unlikely to give their edge away to you. All the features that you have mentioned are typically considered for such models. You may want to consider the orderbook as well.

Not sure if you really meant it, but CL is a very interesting choice for market making. It is liquid but the top of the book is typically very thin and moves a lot. Lack of order by order data from CME makes queue estimation a bit messy as well.

Feel free to send me an email if you want to discuss in private.

The dark is light enough.

kristin


Total Posts: 7
Joined: Sep 2013
 
Posted: 2014-02-04 19:39
Market making is a multi-dimensional problem with a lot of trade-offs and to make money in the most popular products you need to get almost everything right all at once. Simply having a good model is not enough. Execution matters just as much and not just in terms of latency. You have a tough road ahead of you since the guys trading these products now had the luxury of optimizing their strategies when competition was less of an issue and markets were more forgiving.
Previous Thread :: Next Thread 
Page 1 of 1