Forums  > Trading  > Optimal market making in mutiple instruments  
     
Page 1 of 1
Display using:  

gaj


Total Posts: 25
Joined: Apr 2018
 
Posted: 2018-06-18 04:52
How do you skew your quotes when market making multiple instruments? Say you get filled in one instrument and you try to offload your risk in other instruments. So you skew your quotes in other instruments. But which ones and by how much? If you start skewing aggressively across all instruments, then you run the risk of getting filled in all of them. So you have to selectively choose which instruments to skew and not skew too aggressively. How do you guys do this in practice?

tabris


Total Posts: 1256
Joined: Feb 2005
 
Posted: 2018-06-18 06:43
Too many variables and factors. You probably need to be more specific or if you are just asking in general then generally most folks actually skew it based on their probabilities of getting filled conditional on the instrument you just traded.

Dilbert: Why does it seem as though I am the only honest guy on earth? Dogbert: Your type tends not to reproduce.

katastrofa


Total Posts: 458
Joined: Jul 2008
 
Posted: 2018-06-18 08:40
It's driven by risk management. If are long on financials and don't want to get longer, lower you bid quotes for companies in this sector.

ronin


Total Posts: 361
Joined: May 2006
 
Posted: 2018-06-18 15:03

Any time you are quoting multiple instruments, you run the risk of getting filled on all of them. Skewing quotes doesn't change that.

In market making you typically work with some benchmarks. It might be time elapsed ("I want 100 shares in the next 6 minutes"), it might be quantity ("I want 1% of the next 10,000 shares"), it might be price ("I want 100 shares for less than 90.84"), it might be anything else.

If you fall behind your benchmark, you go more aggressive ("10% below benchmark, move 1 level in"). If you are ahead of the benchmark, you go more passive ("10% ahead of benchmark, move one level out").

And then make sure your benchmark for any specific symbol tracks your progression towards net and gross limits.

I am simplifying, but that is the basic idea.


"There is a SIX am?" -- Arthur

gaj


Total Posts: 25
Joined: Apr 2018
 
Posted: 2018-06-18 16:32
I'll be more specific. I was thinking of market making instruments where we have estimated "fair prices". Think of options for example. Say my bid is hit on 40D call. Then I try to hedge my vega by selling other strikes. I'm willing to lose x dollars to hedge this. So I lower my offers in 30D call and 25D call. But if I get filled in both, I'm going to lose 2*x dollars, and I'll have the opposite exposure.

As a manual trader, I would look for the instrument where it is cheapest to hedge, i.e., the one where my "fair ask" is closest to the market. Just focus the quote skewing on this one instrument. I think this is what tabris was saying as well. Is this what people do in practice? It seems a bit troublesome to automate. You have to keep track of which instrument is the cheapest to hedge in real-time and keep readjusting the skew every time it changes.

tabris


Total Posts: 1256
Joined: Feb 2005
 
Posted: 2018-06-19 02:40
That's basically what I'd do. Automating the skew (or just cross the bid/ask to offload the risk like katastrofa) is not really an issue especially if you have the probabilities. Also not all instruments have a high correlation to the product you just traded. Some are stickier than others. The beauty of being a manual trader (I am guessing you are market making at a bank possibly OTC) is you also get the client ID vs probability of you making/losing money if you hedged aggressively or less aggresively. This should definitely help you, otherwise you might not get enough useful information on this and will have to extrapolate.

Dilbert: Why does it seem as though I am the only honest guy on earth? Dogbert: Your type tends not to reproduce.

Jurassic


Total Posts: 161
Joined: Mar 2018
 
Posted: 2018-06-19 22:41
> generally most folks actually skew it based on their probabilities of getting filled conditional on the instrument you just traded.

@tabris how do you get the probabilities?

tabris


Total Posts: 1256
Joined: Feb 2005
 
Posted: 2018-06-20 01:38
if you are market making you would hopefully have had the data to estimate it quantitatively from your empiricals...

Dilbert: Why does it seem as though I am the only honest guy on earth? Dogbert: Your type tends not to reproduce.

Strange


Total Posts: 1450
Joined: Jun 2004
 
Posted: 2018-06-20 03:24
>> As a manual trader, I would look for the instrument where it is cheapest to hedge, i.e., the one where my "fair ask" is closest to the market. Just focus the quote skewing on this one instrument. I think this is what tabris was saying as well. Is this what people do in practice? It seems a bit troublesome to automate. You have to keep track of which instrument is the cheapest to hedge in real-time and keep readjusting the skew every time it changes.

I assume you trade FX vol and are one of the people that rapes me on a regular basis? :) As a manual trader, you almost always end up considering pnl vs effort, e.g. "Strange lifted me in 100m of 1m 40 delta calls. I'll work equal vega in the 25 delta calls at fair offer for the next hour and will cross the spread if I can't get lifted". You got things to see and people to do.

As an automated OMM, you don't have to. You model/system can continuously adjust your skews and attenuate the book. E.g. you offer in both 25 and 30 delta but in modest size. As you are getting filled, (or not) you keep deciding if you still like this exposure, this level of aggression and the instruments you are skewing. Iin fact, IRL, you probably going to better offer smallish amounts across the whole surface (and possibly in related assets, if you are into that type of thing), with some parametric form for the skew vs your current exposures.

The only time you kinda have to do something aggressive is when you get negatively selected and are willing to pay up to get flat. As an automated OMM, most of your efforts are spent to avoid that type of situation.

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

Jurassic


Total Posts: 161
Joined: Mar 2018
 
Posted: 2018-06-20 10:02
@Strange I really cant see why there would be a substantial difference between a human market maker and a OMM in strategy. Why cant the human also skew various other instruments but in modest size?

Jurassic


Total Posts: 161
Joined: Mar 2018
 
Posted: 2018-06-20 11:30
> if you are market making you would hopefully have had the data to estimate it quantitatively from your empiricals...

yes but what does the probability represent, probability of getting filled in a certain amount of time? How would you choose the time period?

ronin


Total Posts: 361
Joined: May 2006
 
Posted: 2018-06-20 12:14
> Why cant the human also skew various other instruments but in modest size?

The machine can monitor the orders and adjust them as appropriate. The human can not, if he is trading more than one or two symbols at the time. So the machine will do a bit extra work for an extra pip or two.

> yes but what does the probability represent, probability of getting filled in a certain amount of time? How would you choose the time period?

Benchmarks my dear boy, benchmarks. There is a reason why you are quoting and ready to take a fill. Think about it as time value of money. The risk that you have in that position is costing you something, and you need to be making at least that to break even. And you are not in business to break even. That translates to a benchmark. And your benchmark may be telling you something like "once I get filled, either I make x pip in the next xyz seconds, or I scratch". So you estimate the probability of getting your x pip fill in the next xyz seconds, and based on that you make some decisions.


"There is a SIX am?" -- Arthur

Jurassic


Total Posts: 161
Joined: Mar 2018
 
Posted: 2018-06-20 12:18
@ronin thanks

> So the machine will do a bit extra work for an extra pip or two.

So what a human market maker better at?

Patrik
Founding Member

Total Posts: 1355
Joined: Mar 2004
 
Posted: 2018-06-20 23:29
There's always a human involved - machines don't construct and run themselves (yet). I think your question is what is a "manual" market maker better at.

The obvious type of situation an automated MM is not very useful is when I call up and want a 2-way in say a 5y out 25d risk reversal in API2 coal, in say around 5mm vega of each strike. In other words illiquid longer term market making where you need to take a lot of basis risk and warehouse substantial risk for significant time. You ain't going home "flat" in any true sense of the word..

That sets your continuum - squeezing the last few bps out of a very liquid underlying with many active strikes is clearly a game for automation, fast execution etc. Big margin, high risk long term illiquid stuff without very active markets and automation buys you "nothing".

Capital Structure Demolition LLC Radiation

Strange


Total Posts: 1450
Joined: Jun 2004
 
Posted: 2018-06-21 01:58
> So what a human market maker better at?
Humans are good (well, better than computers) at making complex risk management decisions. Even as a systematic trader (AMM or otherwise), you will end up making a lot of discretionary decisions. Either because you got stuck with some position and can't get out at a reasonable cost, your model is misbehaving, you need to decide if you going to scale up or scale down etc.

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

TonyC
Nuclear Energy Trader

Total Posts: 1280
Joined: May 2004
 
Posted: 2018-06-21 03:08
> 2-way in say a 5y out 25d risk reversal in API2 coal,

Patrik, i had no idea you were such a BTU Baron

flaneur/boulevardier/remittance man/energy trader

EspressoLover


Total Posts: 337
Joined: Jan 2015
 
Posted: 2018-06-21 09:42
Keep in mind that there's three separate reasons to avoid risk, each with their own motivations and logic. 1) Because you're risk-averse, i.e. you want less variance in PnL. 2) Because of adverse selection, i.e. when you take on risk the market tends to move in the opposite direction. So more risk, equals less PnL. 3) Because it consumes working capital. At one point if you take on enough risk, you will not be allowed to take on more with the money on hand.

A lot of market makers run 10+ Sharpe returns. In which case, 1) is effectively irrelevant. It's also most related to what you're asking about. But many market makers simply do not care about risk management, in the sense of lowering day-to-day volatility in returns. Their concern is much more focused on scalability and capacity, so they want to do every ex-ante profitable trade possible, regardless of its marginal contribution to risk.

From the perspective of 3), things are relatively simple. You have total portfolio size/beta/notional limits. When you quote, it consumes pushes you further towards those limits. At certain times you may want to quote more than the room you have left in your portfolio limits. In which case, whether it's single or multi instrument, the logic's simple. Prioritize the quotes you expect to have the highest profitability.

For 2), in the context of your question, adverse selection can happen in two flavors. The first is instrument specific adversity. In this case the market making sizing can be considered in isolation between instruments. The second is adversity in common factors between instruments. Like maybe you're worried that all your bids might get lifted, just as the entire market is tanking.

But this isn't that much more complex than single instrument sizing. You identify the common factor. Then, just like single instrument sizing, you derive the amount you're willing to quote in the face of that adversity. That gives you a total sizing to budget across the universe. You limit your individual instrument quote sizes accordingly. Like 3), you prioritize the most ex-ante profitable opportunities first.

How much this common-factor adverse selection hurts you depends a lot on your alphas and latency. A big driver for equities market makers is price discovery on index futures. If ES ticks down, a whole bunch of bids on single-names will get swiped all at the same time. But if Alice has a microwave link to Chicago, you can likely cancel your quotes before that happens. In which case, Alice would have much less worry about market-wide adverse selection than Bob.

Good questions outrank easy answers. -Paul Samuelson

ronin


Total Posts: 361
Joined: May 2006
 
Posted: 2018-06-21 11:40



Sorry, couldn't resist...



"There is a SIX am?" -- Arthur

gaj


Total Posts: 25
Joined: Apr 2018
 
Posted: 2018-06-21 13:04
It's great to learn a lot of different perspectives.

The context of my question is actually in automated market making. I normally assign a hedge instrument for every risk factor. So, using katastrofa's example, if I am long financials I will work my hedge in a liquid financial ETF. On the other hand, when I trade manually, I could be smarter and find other instruments that are cheaper to hedge. So I'm trying to translate this manual thinking into the machine. I have experimented a little bit with skewing quotes across all instruments. There's some excellent ideas in this thread regarding sizing and skewing individual instruments.

EL, Your remark about 10+ Sharpe MMs not caring about risk management is surprising to me. Is that due to diversification? I was under the opposite impression that these people try to stay flat as much as possible, so they would puke at the first sign of trouble.

Jurassic


Total Posts: 161
Joined: Mar 2018
 
Posted: 2018-06-21 19:42
do you consider common risk factors only for OMM or as well for d1 products MMing?

Strange


Total Posts: 1450
Joined: Jun 2004
 
Posted: 2018-06-21 23:18
>> MMs not caring about risk management
They sure care about risk management, however they don't care about variance of PnL as a risk management metric.

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

EspressoLover


Total Posts: 337
Joined: Jan 2015
 
Posted: 2018-06-22 19:39
@gaj

A firm like Virtu has positive trading PnL 99% of its trading days. Virtu's biggest risk isn't daily trading volatility, it's having insufficient trading revenue to cover their fixed expenses. (Well, and also blowup risk, but more on that below.) If Virtu was offered the ability to double their trading revenue, keep their fixed expenses constant, but lose money 20% of days, they'd take it in a heartbeat.

The reason a market maker like this tries to stay flat isn't to minimize PnL volatility, it's because of adverse selection. When you're a market maker you can basically only interact with the natural incoming flow. If you're at capacity, then doubling a portfolio's position sizing also means doubling the holding times. And because you're taking the other side of the direction that people want to trade, the market tends to drift against your positions.

The more inventory you hold, the worse this hurts. At some point a marginal increase in risk costs you more in drift then you make in spread+alpha. Hence being proactive about keeping flat or near flat.

A secondary consideration is minimizing the possible damage of a blowup. You can have all the circuit breakers and risk checks you want, but the best risk management has always been a thinly capitalized LLC. If you keep positions small, then that also means you can get away with keeping the account small. If shit hits the fan, you can't lose more money then you put up. (Absent your counterparties deciding to pursue expensive and onerous litigation)

Good questions outrank easy answers. -Paul Samuelson

gaj


Total Posts: 25
Joined: Apr 2018
 
Posted: 2018-06-24 17:21
Thanks, all great points. Although I would argue you still need to manage your inventory risk even without adverse selection, otherwise your inventory will grow to infinity while your PnL stays bounded.

Jurassic


Total Posts: 161
Joined: Mar 2018
 
Posted: 2018-06-29 14:50
@gaj how do you backtest all of this?
Previous Thread :: Next Thread 
Page 1 of 1