Credit trading is changing – does your firm have the data, analytics and technology required to take advantage of this new environment? In this Q&A with Dmitry Pugachevsky, Director, Research, Quantifi, we explore the new requirements for credit analytics, advanced analytics for convertibles, building a credit curve from credit instruments and the challenges arising from IBOR replacement.
We are entering a new era with record issuance and with new types of trading – what does it mean for credit analytics?
As Head of Research at Quantifi, I think there are some high level requirements for modern credit analytics. These are modern requirements, this is not your grandmother’s or mother’s credit analytics, or even back when I was working at Bear Stearns 20 years ago. Firstly, let us start with comprehensive product coverage – this includes bonds, loans, single name CDS, indices etc. Within these bonds, you can have government, corporate, callable, convertible and hybrid bonds – the list goes on. The broader your product coverage, the better your analytics, and the more opportunities you create for your users to trade.
Secondly, firms require an advanced pricing library. This includes multi-factor models, especially for convertibles – covering interest rate, credit, equities, and for hybrid bonds. For relative value analysis, firms need to be able to compare bonds with CDS, imply credit curves from bonds and loans, and calculate sensitivities based on implied credit spreads. Firms are now experiencing the added pressure to address multi-curve and IBOR replacement issues.
Nowadays bond analytics does not exclusively refer to valuation; it should support trading, pricing and structuring, risk and PnL, even VaR. Of course, this functionality should be very efficient, with high performance now becoming one of the main requirements for analytics.
With convertibles, we have three sources of risk: interest rate, credit and equity. If you want to use two factor models, then you can build a two-factor tree – this should be IR-Equity or Equity-Credit. Of course, ideally, you would like all three of them, but three factor trees are very slow. As a result, you have to make a decision; which of the three components are most important?
Dmitry Pugachevsky, Director, Research, Quantifi
Can you elaborate on the analytics required for convertibles?
With convertibles, we have three sources of risk: interest rate, credit and equity. If you want to use two factor models, then you can build a two-factor tree – this should be IR-Equity or Equity-Credit. Of course, ideally, you would like all three of them, but three factor trees are very slow. As a result, you have to make a decision; which of the three components are most important? Which volatilities are most important? Which correlations are most important?
We then need to consider the treatment of equity. Previously, firms were just inputting the spot and the discount rate for equity to grow, but now we have to take into account dividends, a bond’s dividend yield, discrete dividends, the equity funding rate etc. The most direct way is just to input the equity forward curve, which already has all of this information and provides forwards for any given time. Another important factor, which is also quite new in the analytics world – applying no-default probability to the equity forward. By doing so, we consider equity under a no-default assumption because if a default happens, the equity goes to zero and the convertible option is not exercised. Consequently, equity grows not only at the funding rate, but also at funding rate plus credit – this is important to model.
Credit can be treated as either random or non-random, it should be a separate input or should be calibrated to market quote – this flexibility is necessary to imply credit spreads. Everyone knows that convertibles require the ability to handle combination of calls, puts and soft calls, each with their own strike and schedule. It is important to model how this all works, and to allow flexibility for all of these inputs. Prices of convertibles bonds tell you something about credit component, so you want to imply OAS or CDS spread from a market quote and then maybe hedge it with a single name CDS or do relative value trading.
In order to take all bonds into account, you will have to bootstrap and build your credit curve step by step. Thus, a one-year bond will provide credit information from today for one year, and a three-year bond will tell you about credit between one year and three years.
And what about building a credit curve from credit instruments?
Whilst one could imply credit spreads from a bond or a loan, these spreads cannot be used for building a credit curve because they were all defined separately and do not take term structure into account. Each of these spreads characterizes its own bond and its own maturity. For example, if you take a one-year bond and a three-year bond, then you will get a one-year spread and a three-year spread. That three-year spread will not include any information about what the one-year spread told you. In order to take all bonds into account, you will have to bootstrap and build your credit curve step by step. Thus, a one-year bond will provide credit information from today for one year, and a three-year bond will tell you about credit between one year and three years. One should build a curve from the whole set of credit instruments at once. If they have distinct maturities, this can be done using bootstrapping i.e. solving for each maturity one-by-one. If they are not very close, then you can combine bonds, loans, CDS etc.
But what if these loans or bonds are callable?
Then it is not that simple, because for callable bonds the maturity is not that easily defined. If a bond is really in the money, then it will probably be called earlier so its maturity is shorter. Therefore, the information this bond will give you is not about credit to maturity, but rather about credit to the time when it will be called. This is pretty challenging to model when you build the credit curve and will require the use of a multi-dimensional optimizer.
What kind of analytics challenges does the IBOR replacement present?
Around 10 years ago people started using OIS for discounting, and that already created challenges for floaters and loans because this requires multi-curve environment with separation of projection and discounting curves. Now with IBOR replacement on the horizon, new challenges arise. The Alternative Reference Rates Committee (ARRC) recommends that SOFR or Sonia loans and FRNs should be calculated using daily rates and paid in arrears. This creates additional challenges because, unlike swaps, which have a settlement period, FRNs do not have settlement periods, they pay immediately after at the end of the time interval.
Therefore, you do not really have time to calculate your coupon. Since loans and FRNs do not have a payment delay, another mechanism should be used. The ARRC recommends lockouts and lookbacks where you have either shift of time, two to five days, or you freeze your rate for several days. In addition, for lookbacks, you may or may not have an observation shift, and to implement new conventions correctly for FRNs, loans, different countries and different curves, is definitely a major challenge.