Contingent Claims

Introduction

Contingent Claims is a library for modeling financial instruments in Daml. An instrument is represented by a tree of Claims, which describe future cashflows between two parties as well as the conditions under which these cashflows occur.

ContingentClaims.Lifecycle offers lifecycling capabilities, as well as a valuation semantics to map a claim to a mathematical expression that can be used for no-arbitrage pricing.

An example of how to create and lifecycle contracts using Contingent Claims can be found in this tutorial.

In the following we present a user guide for getting started with Contingent Claims-based modeling. It is meant to teach the basics of the framework, but does not cover every aspect. The work is based on the papers [Cit1] and [Cit2], and we recommend that you refer to these for an in-depth understanding of how it works.

The Model

The approach taken in the papers is to model financial instruments by their cashflows. This should be familiar to anyone having taken a course in corporate finance or valuation. Let’s start with an example:

The picture above represents the cashflows of a fixed-rate bond. Or alternatively, you can think of it as a mortgage, from the point of view of the bank: there are interest payments at regular intervals (the small arrows), and a single repayment of the loan at maturity (the big arrow on the right). So how do we go about modelling this?

We use the following data type, slightly simplified from Claim:

data Claim a
 = zero
 | one a
 | give (Claim a)
 | and with lhs: Claim a, rhs: Claim a
 | or with lhs: Claim a, rhs: Claim a
 | scale with k: Date -> Decimal, claim: Claim a
 | when with predicate: Date -> Bool, claim: Claim a
 | anytime with predicate : Date -> Bool, claim: Claim a
 | until with predicate : Date -> Bool, claim: Claim a

There are a couple of things to consider.

First note that the constructors of this data type create a tree structure. The leaf constructors are zero and one a, and the other constructors create branches (observe they call Claim a recursively). The constructors are just functions, and can be combined to produce complex cashflows. For example, to represent the above bond, we could write the following:

when (time == t_0) (scale (pure coupon) (one “USD”)) `and` ...

Let’s look at the constructors used in the above expression in more detail:

• one "USD" means that the acquirer of the contract receives one unit of the asset, parametrised by a, immediately. In this case we use a 3-letter ISO code to represent a currency, but you can use your own type to represent any asset.
• scale (pure coupon) modifies the magnitude of the arrow in the diagram. For example, in the diagram, the big arrow would have a distinct scale factor from the small arrows. In our example, the scale factor is constant: pure coupon = const coupon, however, it’s possible to have a scale factor that depends on an unobserved value, such as a stock price, the weather, or any other measurable quantity.
• when (time == t_0) tells us where along the x-axis the arrow is placed, i.e., it modifies the point in time when the claim is acquired. The convention is that this must be the first instant that the predicate (time == t_0 in this case) is true. In our example it is a point, but again, we could have used an expression with an unknown quantity, for example spotPrice > pure k, and it would trigger the first instant that the expression becomes true.
• and is used to chain multiple expressions together. Remember that in the data definition above, each constructor is a function: and : Claim a -> Claim a -> Claim a. We use the Daml backtick syntax to write and as an infix operator, for legibility.

Additionally, there are several constructors which were not used in the above example:

• zero, used to indicate an absence of obligations. While it may not make sense to create a zero claim, it could, for example, result from applying a function on a tree of claims.
• give would flip the direction of the arrows in our diagram. For example, in a swap we could use give to distinguish the received/paid legs.
• or is used to give the bearer the right to choose between two different claims. This is typically used for options.
• anytime is like when, except it allows the bearer to choose (vs. no choice) acquisition within a region (or timeframe), vs. a specific point in time.
• until is used to adjust the expiration (horizon in [Cit1]) of a claim. Typically, it is used with anytime to limit aforesaid acquisition region.

The tree produced by our expression (corresponding to the cashflow figure above) looks like this:

Composition and Extensibility

Although we could model every subsequent arrow the way we did the first one, as good programmers we wish to avoid repeating ourselves. Hence, we could write functions to re-use subexpressions of the tree. But which parts should we factor out? It turns out that Finance 101 comes to the rescue again. Fixed income practitioners will typically model a fixed-rate bond as a sum of zero-coupon bonds. That’s how we model them in Claims.Util.Builders. Below are slightly simplified versions:

zcb maturity principal asset =
 when (time == maturity) (scale (O.pure principal) (one asset))

Here we’ve just wrapped our expression from the previous section in a function zcb, that we can reuse to build a fixed-rate bond:

fixed : Decimal -> Decimal -> a -> [Date] -> Claim a
fixed principal coupon asset [] = zero
fixed principal coupon asset [maturity] =
  zcb maturity coupon asset `and` zcb maturity principal asset
fixed principal coupon asset (t :: ts) =
  zcb t coupon asset `and` fixed principal coupon asset ts

We define the fixed rate bond by induction, iterating over a list of dates [t], and producing multiple zero-coupon bonds zcb combined together with and:

• The first definition covers the trivial case where we pass an empty list of dates.
• The second definition handles the base case, at maturity: we create both a coupon (interest) payment, and the principal payment.
• The third definition is the induction step; it peels the first element off the list, and calls itself recursively on the tail of the list, until it reaches the base case at maturity.

This re-use of code is prevalent throughout the library. It’s great as it mirrors how instruments are defined in the industry. Let’s look at yet another example, a fixed vs floating USD/EUR swap.

type Ccy = Text
usdVsEur : [Date] -> Claim Ccy
usdVsEur =
   fixed 100.0 0.1 "USD" `swap` floating (spot "EURUSD" * pure 100.0) (spot "EURUSD") "EUR"

We define it in terms of its two legs, fixed and floating, which themselves are functions. We use swap in infix form, and partially apply it - it takes a final [Date] argument which we omit, hence the resulting signature [Date] -> Claim Ccy.

As you can see, not only is this approach highly composable, but it also mirrors the way derivative instruments are modelled in finance.

Another major advantage of this approach is its extensibility. Unlike a traditional approach, where we might in an object-oriented language represent different instruments as classes, in the cashflow approach, we do not need to enumerate possible asset classes/instruments a priori. This is especially relevant in a distributed setting, where parties must execute the same code, i.e., have the same *.dars on their ledger to interact. In other words, party A can issue a new instrument, or even write a new combinator function that is in a private *.dar, while being able to trade with party B, who has no knowledge of this new *.dar.

Concerning Type Parameters

The curious reader may have noticed that the signature we gave for data Claim is not quite what is in the library, where we have data Claim t x a o. In our examples, we have specialised this to type Claim' t x a o = Claim Date Decimal a a. Parametrising these variables allows us to reason about Assets and Observations that appear inClaims as function-like objects. The main use of this is to create claims with ‘placeholders’ for actual parameters, that can later be ‘filled in’ by mapping over them (mapParams).

The Time Parameter

t is used to represent the first input argument to an Observation, and above we used Date for this purpose. One reason this has been left parametrised is to be able to distinguish different calendar and day count conventions at the type level. This is quite a technical topic, but it suffices to know that for financial calculations, interest is not always accrued the same way, nor is settlement possible every day, as this depends on local jurisdictions or market conventions. Having different types makes this explicit at the instrument level.

Another use for this is expressing time as an ordinal values, representing e.g. days from issue. Such a Claim can be used repeatedly to represent instruments issued at different dates, but with the same durations. For example, consider a series of listed futures or options which are issued with quarterly/monthly maturities - their duration is about the same, but they are issued on different dates.

The Asset Parameter

a, as we already explained, is the type used to represent assets in your application. Keeping this generic means the library can be used with any asset representation. For example, you could use one of the instrument implementations in Daml Finance, but are not forced to do so.

The Observation Parameter

o is the type used to represent Observations, which are time-dependent quantities that can be observed at any given time (such as the “EURUSD” exchange rate in the example above).

The Value Parameter

x is the ‘output’ type of an Observation, but it can also serve as input when defining a constant observation using, e.g., Observation.pure 10.08.

Lifecycling

So far we’ve learned how to model arbitrary financial instruments by representing them as trees of cashflows. We’ve seen that these trees can be constructed using the type constructors of data Claim, and that they can be factored into more complex building blocks using function composition. But now that we have these trees, what can we do with them?

The original paper [Cit1] focuses on using these trees for valuing the instruments they represent, i.e., finding the ‘fair price’ that one should pay for these cashflows. Instead, we’ll focus here on a different use case: the lifecycling (aka safekeeping, processing corporate actions) of these instruments.

Let’s go back to our fixed-rate bond example, above. We want to process the coupon payments. There is a function in `Lifecycle.daml <./daml/ContingentClaims/Lifecycle.daml>`__ for doing just this:

type C t a o = Claim t Decimal a o

-- | Used to specify pending payments.
data Pending t a = Pending with
  t : t
    -- ^ Payment time.
  amount : Decimal
    -- ^ Amount of asset to be paid.
  asset : a
    -- ^ Asset in which the payment is denominated.
    deriving (Eq, Show)

-- | Returned from a `lifecycle` operation.
data Result t a o = Result with
  pending : [Pending t a]
    -- ^ Payments requiring settlement.
  remaining : C t a o
    -- ^ The tree after lifecycled branches have been pruned.
    deriving (Eq, Show)

-- | Collect claims falling due into a list, and return the tree with those nodes pruned.
-- `m` will typically be `Update`. It is parametrised so it can be run in a `Script`. The first
-- argument is used to lookup the value of any `Observables`. Returns the pruned tree + pending
-- settlements up to the provided market time.
lifecycle : (Ord t, Eq a, CanAbort m)
  => (o -> t -> m Decimal)
  -- ^ Function to evaluate observables.
  -> C t a o
  -- ^ The input claim.
  -> t
  -- ^ The input claim's acquisition time.
  -> t
  -- ^ The current market time. This is the time up to which observations are known.
  -> m (Result t a o)

This may look daunting, but let’s look at an example in `ContingentClaims/Test/FinancialContract.daml to see this in action:

        do
          t <- toDateUTC <$> getTime
          let
            getSpotRate isin t = do
              (_, Quote{close}) <- fetchByKey (isin, t, bearer)
              pure close
          lifecycleResult <- Lifecycle.lifecycle getSpotRate claims acquisitionTime t

The first argument to lifecycle, getSpotRate, is a function taking an ISIN (security) code, and today’s date. All this does is fetch a contract from the ledger that is keyed by these two values, and extract the price of the security.

The last two arguments are simply the claims we wish to process, and today’s date, evaluated using getTime.

The return value, lifecycleResult, will contain both the remaining tree after lifecycling, and any assets that need to be settled. In our running bond example, we would extract the coupon from the first payment, and return it, along with the rest of the tree, after that branch has been pruned (depicted greyed-out below):

You may wonder why we’ve separated the settlement procedure from the lifecycling function. The reason is that we can’t assume that settlement will happen on-chain; if it does, that is great, as we can embed this call into a template choice, and lifecycle and settle atomically. However, in the case where settlement must happen off-chain, there’s no way to to do this in one step. This design supports both choices.

Pricing (Experimental)

This is an experimental feature. Expect breaking changes.

The ContigentClaims.Valuation.Stochastic module can be used for valuation. There is a fapf function which is used to derive a fundamental asset pricing formula for an arbitrary Claim tree. The resulting AST is represented by Expr, but can be rendered as XML/MathML with the provided MathML.presentation function, for display in a web browser. See the Test/Pricing module for examples. Here is a sample rendering of a margrabe option:

<math display="block"><msub><mi>USD</mi><mi>t</mi></msub><mo>⁢</mo><mo>𝔼</mo><mo>⁡</mo><mrow><mo fence="true">[</mo><mrow><mo fence="true">(</mo><msub><mo>I</mo><mrow><msub><mi>AMZN</mi><mi>T</mi></msub><mo>-</mo><msub><mi>APPL</mi><mi>T</mi></msub><mo>≤</mo><mn>0.0</mn></mrow></msub><mo>⁢</mo><mrow><mo fence="true">(</mo><msub><mi>AMZN</mi><mi>T</mi></msub><mo>-</mo><msub><mi>APPL</mi><mi>T</mi></msub><mo fence="true">)</mo></mrow><mo>+</mo><msub><mo>I</mo><mrow><mn>0.0</mn><mo>≤</mo><msub><mi>AMZN</mi><mi>T</mi></msub><mo>-</mo><msub><mi>APPL</mi><mi>T</mi></msub></mrow></msub><mo>×</mo><mn>0.0</mn><mo fence="true">)</mo></mrow><mo>⁢</mo><msup><mrow><msub><mi>USD</mi><mi>T</mi></msub></mrow><mrow><mo>-</mo><mn>1.0</mn></mrow></msup><mo>|</mo><msub><mo mathvariant="script">F</mo><mi>t</mi></msub><mo fence="true">]</mo></mrow></math>

You can cut-and-paste this into a web page in ‘developer mode’ in any modern browser.

References

[Cit1]

(1, 2, 3) Jones, S. Peyton, Jean-Marc Eber, and Julian Seward. “Composing contracts: an adventure in financial engineering.” ACM SIG-PLAN Notices 35.9 (2000): 280-292.

[Cit2]

Jones, SL Peyton, and J. M. Eber. “How to write a financial contract”, volume “Fun Of Programming” of “Cornerstones of Computing.” (2005).

The papers can be downloaded from Microsoft Research.