Illiquid CDS Spread Modeling

Back in the days when pricing a CDO/Baskets/LCDS were hot and perhaps get you a quant job doing these type of research. Days when Lehman, Merrill, and Bear publish research papers about credit derivatives pricing with armies of quants behind the calibrating the results and trying to help us intuitively understand the concept while accepting the assumptions could be garage in and garage out. I had my contributions in this as well ending up as a standard 6 hour courses in financial engineering programs. Pricing CDO/Basket Products course by David Lu

This was another article co-authored with my classmate and professor. This was after the crisis where try to crack the pricing of a Loan Credit Default Swap Ong_Li_Lu_LCDS Journal of Credit Risk

Interesting “a-ha” moments in life may many times happened after a “signal” made by a person or event. I started to think about reinforcement learning in 2005 when I first started to think about robust portfolio selection. On a side note, Andrew Ng wrote that he started to think about optimal control in his Phd Thesis back in 1998. We will talk about relationships in optimal control and reinforcement learning in later posts.

So when did I actually thought about supervised learning or machine learning? It happened in 2009 when I actually interviewed for a position at large rival investment bank at time square.  The interviewer asked me how would I handle illiquid CDS spread. I told him I’d map to a liquid CDS curve. How would you price the curve? I told him I would map it with industry sector, region, ratings, capital structure, recovery rate or to an index. He didn’t seemed satisfied and I didn’t get the job. After the interview, though it was a front office COO role on the description, I notice he was a quant. As much as I made the mistake of not looking him up on Linkedin beforehand, it had me thinking about what he wanted was my thoughts on how I am going to model this spread.

We can estimate illiquid CDS spreads, or in other words to estimate unobserved CDS spreads for companies with illiquid debt based on observed spreads and other characteristics of other, liquid companies, we can use supervised learning or classification methods.

Our inputs can be the X features for the company: ratings, industry sector, geographical sector,financial ratios, implied volatilities, and Moody’s EDF (Expected Default Frequency). Take these observable features and train for an output for the observed universe of liquid CDS. We can use Decision Trees, Support Vector Machine or Neural Networks to handle this problem. It turns out that support vector machines are pretty good at handling these type of problems. If the readers interested of these type of problems it is also used in internal ratings modeling where this can be applied in regulatory issues such as Basel which the rules wouldn’t allow you to use rating agencies ratings.



Machine Learning Applications in Global Markets and Risk Management

First thing everyone would think about is definitely algorithmic trading, in fact that’s what mostly we’ll talk about in this forum. However, there are many other fields there I’ve either seen or implemented myself. Econometric predictive models in balance sheet management or regulatory CCAR (Comprehensive Capital Analysis Review) processes. Though mostly is regarding simple techniques such as linear regression. Other usages include regime change detection models stock segmentation etc. You may many of cases that Finance or Legal departments use text processing or document data mining. Consumer banking  had many techniques applied to customer segmentation and default modeling in credit cards. While we aim a lot of effort in algorithmic trading, Risk Management and regulatory areas had a lot to explore and potential implementations. Some of the examples included mortgage segmentation, prepayment rates, internal rating modeling, probability of default modeling, Loss Given Default modeling, and other segmentation problems.

How Quants Evolve in Global Markets

This dates back before 2000, when many Physics and Maths Phds graduates or Professors rush into wall street to become quants. Here is one of the books I was reading before deciding to come to the states for Financial Engineering in 2004.

Those time most folks do either interest rate modelling or prepayment models for mortgage back securities.

Year 2000 was a shift that quants starting to realize, hey I can do a somewhat Black Scholes like model to price collateralized debt obligations, a structured financial product backed by a pool of loans. It was initiated by David Li,

The idea was actually circle around in 1987 from KMV’s V – Oldrich Vasicek with a paper named probability of Loss on Loan Portfolio. Though David was once called the world’s most influential actuary, model itself was called the “recipe for disaster” after the global financial crisis between 2008 and 2009. Now of course, between 2000 and 2008, there were many variations of the model, Random Factor Loading, Levy Process to name a few.

After 2009, with the push of many regulatory reforms, people start to dive into pricing adjustments such as credit valuation adjustments, funding valuation adjustments, margin valuation adjustments, and capital valuation adjustments. We later on call these XVAs. Many of the works are surrounding either counterparty risk, liquidity, and how do we model the denominator of the capital ratio, aka risk weighted assets.

At the mean time, in the “blackbox” world, the usage of algo trading expands rather quickly throughout the years. See Michael Lewis’s “Flash Boys.” Fiction or not, High Frequency Trading firms and quants continue to figure out ways to beat the market while hiding in the scenes.

We will describe some of the machine learning techniques to reverse engineer these strategies and also risk management aspects on these strategies



Optimal Portfolio Selection Using Linear Matrix Inequilities

In Optimal Control Theory, there are many researcher who consider robustness more than performance, which is how we deal with uncertainty. There is a mathematically method as describe in the research of controlling high rise building that we could use to solve for the optimal objective. We introduce Linear Matrix Inequilities. There are many researchers and pioneers we specialized in this theory for instance here:

Here is the early work we did back in 2005. Robust Portfolio Selection The forecasting of time can actually be implemented in Recurrent Neural Networks or Long Short Term Memory as of today. We will discuss that further in later posts along with recurrent reinforcement learning.

Controlling High Rise Buildings

I was very fortunate in a sense that I get to work on a project at year 2000 which is all about heavy lifting research with no sights of implementing it in reality.  We had couple sets of sensors, a wind tunnel not quite the $10 million Ferrari Wind Tunnel that tests out 0.2 coefficient of drag; a model high rise building and bunch of little wooden bricks to model a city. The goal was to put an actuator on the top of the building and try to reduce the sidesway and acceleration on each degree of freedom based on the wind excite. Concretely, an optimal control problem, you normally have a state and an objective, we’ll link this back to reinforcement learning later on. Initially, we tried to do Linear Quadratic Gaussian Control, which is adding a Kalman Filter on a Linear Quadratic Regulator controller. My advisor, Dr. J.C. Wu decided that wasn’t fancy enough. I was to come up with a Robust Controller with so called Linear Matrix Inequities to solve for the controller. That was basically what I did the next 2 years. If you are interested, I’ve attached the full story here. Active Control on Wind Excited High Rise Buildings published_2006

On the next post, we’ll talk about how to use this methodology on optimal portfolio selections

A New Beginning

Never had occur to me to write about quantitative stuff or anything work related since the passion should always continue within and never had to be a piece of the memory. Now that I am entering a certain age and an age where some people called midlife, I guess it’s time to start writing and recording what happened or what should of happened. It almost felt like I am writing letters to my kids.

The crisis started with trying to do research on golf. From how the shots are made to how the materials are constructed to fit the feel people like and with all the sense of complexity, how to get lower scores. When you start to get into golf at the time when you start to put salt and pepper on your hair but not when you are 8, something must be terribly wrong.

This site will not be about golf unless I come up with some disrupting devices fairly soon. We’ll be talking more about artificial intelligence, the ‘hype’ created since 2010, and its application to many fields including financial services, health care, advertisement, image  and text processing and not self driving cars; I preferred to drive them myself.