Forums  > Trading  > Algorithmic market maker in "less liquid" OTC IRS  
     
Page 1 of 1
Display using:  

gammaphreak01


Total Posts: 23
Joined: May 2007
 
Posted: 2019-10-21 18:14
Contrasting the problem of creating an algorithmic market maker in OTC IRS vs e.g. FX spot is an interesting problem.

The extension from a scalar (spot FX) to a set of related rates (swap curve) seems pretty difficult.

I am imagining some concepts related to internalisation of "swap curve risk" coming into play

I suppose that market impact is something that should also make an appearance.

I am interested to brainstorm the problem, haven't found much info on the interweb and wondering if there are like minded (or similarly challenged) people on this random forum in this corner of the web who are willing to discuss.

AlexS


Total Posts: 2
Joined: Apr 2012
 
Posted: 2019-10-29 13:36


What use case do you have in mind for this IRS automated market maker? Is it an amm to complement voice trading on a swap desk at an IB (i.e. auto-quoting and RFQ pricing, auto-hedging, e-book management, etc.), or something else?

Deus ex machina

rickyvic


Total Posts: 201
Joined: Jul 2013
 
Posted: 2019-12-09 13:33
Any idea on the state of G10 long term irs electronic liquidity?
I have been away from that market for a long time I was curious.

"amicus Plato sed magis amica Veritas"

gammaphreak01


Total Posts: 23
Joined: May 2007
 
Posted: 2019-12-25 18:06
good question:

Initially to complement voice trading on a swap desk, but to be fair, looking to drive a capability that starts redefining how the desk manages their overall risk. We are a small shop and the voice guys are pretty resistant to "interference" from the quant trading side. We also focus on a semi-liquid EM market where it could be argued that electronic trading should fear to tread - but I have heard that before and seen decent progress made from those with determination and focus.

Not sure of any other threads or phorums where such businesses are discussed.

Practically speaking:
1) managing IRS positions versus more liquid treasury positions using either something like PCA analysis to assist with risk bucketing. Perhaps run some form of mean-reversion in this space

don't know - spitballing a bit here...

gammaphreak01


Total Posts: 23
Joined: May 2007
 
Posted: 2019-12-25 18:08
Anecdotally, I have heard of response times in the ms being required to be relevant in this space on RFQ's (certainly in G3)

AlexS


Total Posts: 2
Joined: Apr 2012
 
Posted: 2019-12-30 16:00
Interesting... In my case (at my previous place of work / the last one on the sell side :-) ) the strongest push-back came not from the voice traders (the desk actually initiated the project) but from the "duration management" people, i.e. dudes who were running their own AMM in OTR UST space, and were providing auto-hedging service to the swap desk.

For me an AMM setup in IR swaps space is essentially an optimal control problem. The duration hedging is done in order-driven markets (ED futures, treasuries, treasury futures order books), the swaps flows are in quote driven market (auto-quoting / RFQ pricing). The rate of hedging in OD markets and pricing in QD market (i.e. B/A spread for RFQ pricing) are the controls.

In practice, the swaps AMM runs two optimization problems: periodic re-hedge (on every trade + market move) and pricing (on every inquiry). The utility is a combo of the swaps book risk, hedging costs and any RV (reflecting the set of strategies you're running). This insures that your RFQ pricing properly reflects your e-book risk, projected hedging costs and RV positioning.

In terms of risk representation (you mentioned PCA), my personal preference is a hierarchical risk: the top level (1) is e.g. 2s, 5s, 10s, bonds (hedge duration, manage spreads), the next level (2) e.g. 4s, 7s and 20s (mean reverting / can be traded), and the following level (3) - all other tenors (as spreads to level (2) / mean-reverting, low VAR). With a bit of linear algebra this can be done in PCA basis. It just boils down to how much mean reversion you get in level (2) curve spreads. My personal choice, is a basis based on a term structure model, e.g. MFC, as it reflects the hierarchical structure of the swap curve by design.

An obvious but important point is that it certainly helps to be big in IR swaps market. Among other things this gives you better estimates for the fill probabilities, required for the RFQ pricing optimization problem (your RFQ table is large, thus, in addition to the trades you won / was second, you can identify the direction / spread for larger population of trades in SDR). Also, since it does not cost you anything to setup RV positions (you may even get paid a bit in b/a spread), with a steady flow of RFQs you have better chances of picking up extra P&L on your RV strategies (either via mean-reversion or just by selling them to your clients :-))

Deus ex machina
Previous Thread :: Next Thread 
Page 1 of 1