Forums  > Basics  > legging a spread is like spreading the legs  
Page 1 of 1
Display using:  


Total Posts: 1677
Joined: Jun 2004
Posted: 2020-04-14 21:28
I am trying to trade a multi-leg spread/basket across different assets. It has both long and short legs, with varying liquidity profiles and different contract sizes. What's the best approach to legging that monster together?

In this specific case (one illiquid short leg and a bunch of liquid long legs), I am thinking that I want to put passive orders on the single illiquid leg and pick up the more liquid stuff aggressively as I am getting filled. However, it would be cool if I can come up with a general approach.

'Progress just means bad things happen faster.’


Total Posts: 474
Joined: Jan 2015
Posted: 2020-04-14 22:53
Here's how I would approach it, but YMMV. First, I'm guessing (but could be mistaken):

O(TCost Illiquid Leg) >> O(TCost Liquid Legs)

In which case the overriding objective is to minimize the t-costs on the most expensive, illiquid leg. And that almost certainly means working the schedule passively and patiently. Like you suggest, let the illiquid execution dictate the cheap legs.

If it costs a lot less to leg in to the liquid instruments, you maybe don't care how they get filled. The relatively de minims difference between simple and efficient execution is maybe not worth extra complexity. But optimal execution of the liquid legs would basically involve trading off the risk of being unhedged against the marginal cost of faster execution.

And that basically involves coming up with a risk budget. How much is X% in additional portfolio volatility worth to save Y% drag on the strategy returns? (I'm assuming the unhedged exposure has zero drift, but if not it's easy to add a term for that.) So, you start by patiently working the illiquid leg. Then immediately after you get that fill, you solve for best execution on the liquid legs. Minimizing X% against Y%.

The biggest wrench to the above approach would be if the t-costs on the liquid legs still makes up a significant proportion of the strategy's aggregate t-costs. (For example VX is very liquid, but still expensive to trade because of the wide tick size.) Then it essentially becomes a 3-body problem, and the only way to solve it is to throw clock cycles at a numerical optimizer.

Particularly if you're dealing with multiple thick-book instruments, you probably want to be working in each leg's respective queue right from the start. But that poses a problem where you'll get filled at different times and have non-deterministic unhedged positions. One way to hack around this is if you can find a "bridge hedge". Some instrument that you temporarily hedge with while the actual legs are getting filled. Maybe it's not the perfect hedge, but its cheap and fast to trade, and does a good enough job for the execution window. For example if you don't want to cross the spread on STIR futures, you can proxy an equivalent duration hedge with ZB for a few hours while you work the Eurodollar queue.

Good questions outrank easy answers. -Paul Samuelson


Total Posts: 1347
Joined: Jun 2005
Posted: 2020-04-15 07:10
by waiting and filling illiquid leg, opening delta (or whatever measure needed to cover) and urging to close it with liquid leg you become liquidity prey in illiquid market.

if I understood you correctly.

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)


Total Posts: 109
Joined: Jul 2018
Posted: 2020-04-15 11:55
@nikol I don't think so. Purely hypothetical example but let's say it's index arb, - you're replicating it with some subset of stocks, some of which might be illiquid. Maybe I need to short one bank stock in particular but can't, so I might be able to temporarily hedge in a financials ETF (which should have similar beta and be cheaper) until I get the passive fill in the individual stock I want.

Silly toy example but you get the idea.

did you use VWAP or triple-reinforced GAN execution?


Total Posts: 1347
Joined: Jun 2005
Posted: 2020-04-15 13:07
Yes, I get it.

My example (and experience of lame beginner) related to A1-A2 simple arb, where price(A1) is function of price(A2), hence A1 order follows A2 as a dog tail.

In your case, even if you try to capture similar spread, you can hedge in any of "similar" markets not only A2, but in A2i, A2ii, A2iii etc. And your price(A1) might be a function of all those A2i...A2xx, so the predator will have a difficulty to bounce your A1 order for cheap.

... What is a man
If his chief good and market of his time
Be but to sleep and feed? (c)

Nuclear Energy Trader

Total Posts: 1361
Joined: May 2004
Posted: 2020-04-15 20:47
I'm a little more curious as to how (once the spread has gone your way) you would take the spread off.

flaneur/boulevardier/remittance man/energy trader


Total Posts: 667
Joined: May 2006
Posted: 2020-04-15 21:36
> I'm a little more curious as to how (what's the spread has gone you are way) you would take the spread off.

Second that. Especially if the liquidity gap is *that* high.

The standard answer from the derivatives world would be that you don't - you run the illiquid leg which is making you money to maturity.

But nobody said it even has a maturity...

"There is a SIX am?" -- Arthur


Total Posts: 117
Joined: Apr 2018
Posted: 2020-04-16 15:03
I would start with a fair value for each leg, it could be as simple as mid price. Then skew each asset according to whether you want to buy or sell. Then you can work your order at fair_value+skew+edge, where edge depends on the asset's liquidity, spread, queue size, etc.

How much to skew is the difficult part. The proper way is to define a value function for your portfolio. Then ask "if I buy 1 lot of this asset, how much does it increase/decrease my value function". The skew should be proportional to this number.


Total Posts: 1677
Joined: Jun 2004
Posted: 2020-04-20 03:48
Thanks a lot guys! I am going to get some liquidity metrics first so I can formulate a better idea and probably share it here.

'Progress just means bad things happen faster.’
Previous Thread :: Next Thread 
Page 1 of 1