Quick Takes: ISDA SIMM – What changes in v2.6?

Quick Takes: ISDA SIMM – What changes in v2.6?

[00:00:00] Ali Curi: Hi everyone, and welcome to ION Markets Quick Takes. I’m Ali Curi, and every week, along with my guests, Amir Khwaja and Chris Barnes, we take a quick dive into the headlines on the Clarus blog. Let’s get started. Hi, Amir. Hi, Chris.

[00:00:16] Amir Khwaja: Hey, Ali.

[00:00:17] Chris Barnes: Hey, Ali. How are you doing?

[00:00:18] Ali Curi: Doing great. Amir, let’s start with you.

[00:00:21] Ali Curi: What are your Quick Takes for this week? Which headline from the Clarus FT blog would you like to discuss?

[00:00:26] Amir Khwaja: Yes, I’m going to cover, ISDA SIMM What Changes in Version 2.6.

[00:00:30] Amir Khwaja: ISDA calibrates a SIMM version each year and that really changes the initial margin that you face with your bilateral derivatives, your counterparties, it goes up and down depending on volatility.

[00:00:42] Amir Khwaja: So in December this year, there’ll be a new version and some asset classes like equity and commodity will have higher margin. Some will be unchanged, some will be lower. So in my blog, I look at what’s changed and how that affects your margin. You currently collect or post to your customers or your counterparties and what will happen post [00:01:00] December 2, 2023.

[00:01:01] Amir Khwaja: And I wanted to give Chris a chance to ask some questions.

[00:01:04] Chris Barnes: Thank you, Amir. Yeah. So I think the big question is what has happened to Sterling risk weights? I imagine the biggest reason for doing 2.6 is to capture all of the volatility of the mini budget.

[00:01:16] Amir Khwaja: So that was actually captured. Prior to an annual recalibration, there was actually an interim recalibration this year for the first time, version 2.5a in July. And that was caused by, Sterling rates in Q4 last year.

[00:01:31] Chris Barnes: So 2.5a covers the meltdown in Sterling markets?

[00:01:36] Amir Khwaja: Yes. In fact, I think because that was so unprecedented, they had to do an interim version for the first time in five years to cater for, primarily the moves in Q4 last year in Sterling.

[00:01:47] Chris Barnes: Okay, so 2.6 is really looking at more of the bonds sell off that as a result in global markets as a result of changes.

[00:01:55] Amir Khwaja: Yeah. So I think 2.6, has kept the same weights as 2.5a for [00:02:00] interest rate moves. A correlation between risk factors and has recalculated all the other asset classes, equity, commodity, credit, et cetera. And those have changed more.

[00:02:08] Chris Barnes: And in relation to how we look at CCP margin is it fair to say it’s all run on historic VAR? I’m not sure there are any CCPs that don’t use it, are there?

[00:02:17] Amir Khwaja: Correct. Yeah, primarily, yes. So I think the big difference with ISDA SIMM, it’s a value at risk methodology, but it’s based on a parametric value at risk, not a structural simulation.

[00:02:27] Amir Khwaja: So that has advantages and disadvantages. Broadly, I think for such a widely adopted industry calculation that everyone needs to implement, it’s better to be simplistic and have it work than to try and get a few more percent accuracy with a full reverse simulation. So I think yes, there are well known pros and cons to the two methodologies.

[00:02:48] Chris Barnes: So do we expect those models to be more stable than CCP models?

[00:02:53] Amir Khwaja: Yeah I guess there’s some big differences. Firstly, the clearinghouses, CCPs, are run initial margin. Every [00:03:00] day, add new historical data to the scenario set. So it takes account of recent market volatility in the model. So IM will change, each day as we get new historical data.

[00:03:11] Amir Khwaja: The ISDA SIMM method is calibrated once a year. So it really, the weights only change once a year. Yeah, so it is less responsive to recent market events.

[00:03:21] Chris Barnes: And when we look at changes, from version 2.5 to version 2.6, are we literally just putting in the risk weights and the correlation factors into a spreadsheet and looking at the changes?

[00:03:35] Chris Barnes: It’s similar to how you’ve done it on the CCP disclosures. If it’s moved by more than 20%, it’s interesting. If it hasn’t, we ignore it.

[00:03:42] Amir Khwaja: Yes. So I start with that because that is a very transparent overview as to what’s changed. But in a real portfolio, it’s hard to get the effect because the effects are not additive by looking at single risk factors.

[00:03:54] Amir Khwaja: So really what we do in CHARM is that we can run your existing portfolio with the new version [00:04:00] and the current version. And that gives you the real change in IM for a counterparty portfolio. Because of aggregation, it’s a nested VAR calculation because of correlations looking at single risk factors gives you an idea, but it doesn’t really make, give you a correct number for a portfolio because you have risk offsets. Actually, and the correlations also change, right? Significantly. So clearly, the higher the correlation, the closer to its factors IM is to their sum and the lower the correlation, the further away it is, right? So I think that has an effect.

[00:04:32] Chris Barnes: I remember writing the original ISDA SIMM blogs on how to build the model in Excel. And when you presented, a) with the documentation, it’s quite daunting, but b) when it’s just a list of risk factors, it looks like quite a simple model to implement. But then when you’re trying to look at changes over what your portfolio look like on these test weights versus what it looks like now, it’s quite tricky, right?

[00:04:56] Chris Barnes: But looking at it from a blog viewpoint, I can’t help [00:05:00] noticing that the output for the blog is quite formulaic. So it makes me start to think, is this a type of blog that you could give BARD or give ChatGPT, “is the same risk weights version 2.5, is the risk weights version 2.6,” and it could generate meaningful output. Do you think that will be a possibility?

[00:05:26] Amir Khwaja: I think it probably would. Yes. I think I could send it given they’re both URLs. I could say, “compare these two documents with their URLs,” and see what it came up with. Yeah, I haven’t tried that, Chris. It will be interesting. I think it will certainly give me a start point to edit.

[00:05:39] Amir Khwaja: The problem is, it doesn’t present, it doesn’t output tables and data at least not the free versions. So the output would have to be worked on a bit, but yes, it would be helpful. It might be a bit too verbose. It’s hard to know. You’d have to really be an expert in prompt engineering and say, “give me X words,” or I think you have to play with the actual interaction with BARD or ChatGPT to exactly what you [00:06:00] wanted.

[00:06:00] Chris Barnes: That’s based on the free versions, right?

[00:06:02] Amir Khwaja: Yes.

[00:06:02] Chris Barnes: If you think about it, what Clarus are doing here with these types of blog is just generating free material for other people to train AIs to write Clarus blogs. Because the output is structured. Because we’re analyzing changes from ISDA SIMM version X to X.

[00:06:19] Amir Khwaja: Ah, good point.

[00:06:20] Chris Barnes: There’s a given output each time, right? And you don’t necessarily even need these new fancy generative AIs to be able to train a model that knows what it’s looking for and can fill into that type of blog process.

[00:06:36] Amir Khwaja: Yeah, that’d be great. It’d make my job easier. I could spend, instead of spending hours, I could spend a fraction of the time on that blog. Yeah. At the moment.

[00:06:44] Chris Barnes: Precisely. We should stop writing and we should start training AIs to write the Clarus blog for us.

[00:06:51] Amir Khwaja: Because, but what I find is, but I find the act of writing makes me, forced me to read something in detail, which aids understanding, right? So that time [00:07:00] spent writing, I’m reading and learning. Without that time, I would know less. The AI might help me, but then I’m reliant on reading the AI output. I feel I would have less understanding.

[00:07:13] Chris Barnes: So in this case, AI will make us more stupid.

[00:07:18] Amir Khwaja: Less in depth. It’ll make us higher level, more superficial in our understanding.

[00:07:23] Chris Barnes: All right, Ali, I think on that conclusion, it’s over to you.

[00:07:26] Ali Curi: Chris and Amir, thank you for sharing your Quick Takes this week. Let’s do it again next week.

[00:07:31] Amir Khwaja: Thanks, Ali.

[00:07:32] Chris Barnes: Thanks, Ali.

[00:07:33] Ali Curi: And that’s our episode for today. You can read more about these topics on the Clarus blog, and you can follow ION Markets on X, formerly Twitter, and on LinkedIn.

[00:07:41] Ali Curi: Until next week, thank you for joining us.