Overview

This document describes the spread model for pricing cancellable swaps referenced against the IPOR index. The objective is to construct a fast, stable and efficient pricing scheme for obtaining rates for cancellable payer and receiver swaps (28 day instruments) in such a way that it can be deployed on the blockchain.

Introduction

The IPOR index is obtained as a composite of AAVE and Compound rates (see [3]). Figure 1 shows the behaviour of the index, which notably has volatile (e.g. 1Q 2021) and quiet (e.g. Jun-Aug 2021) periods.

Figure 1 - IPOR index (2 Dec 2020 to 4 Feb 2022)

Figure 1 - IPOR index (2 Dec 2020 to 4 Feb 2022)

One notable feature of this index is that it exhibits upward shocks, generally followed by rapid downward retracement. This indicates that a random model with Brownian motion is not adequate - upon fitting a normal distribution to the data, it is evident in Figure 2 that the distribution (in blue, with histogram) is considerably more fat-tailed than a Brownian motion model can describe.

Figure 2 - Histogram of observed daily increments in IPOR together with KDE-based empirical density (blue line) and normal density (red line)

Figure 2 - Histogram of observed daily increments in IPOR together with KDE-based empirical density (blue line) and normal density (red line)

Based on this, we argue that we need a model for IPOR rates which captures randomness together with jumps, and the tendency for these jumps to be followed by retracement. Consequently, we adapt the well known Hull-White model [1] with an extra jump term, and use the mean reversion parameter to coerce the model to bring IPOR rates back down after the upward jump shocks.

Hull-White Jump-Diffusion Model

The model used for IPOR rates $r_t$ is

$$ dr_t = \alpha(\theta_t - r_t)\,dt + \sigma_tdW_t + J dN_t\,\! $$

where $\theta_t$ denotes the mean reversion level, $\alpha$ the mean reversion rate, $\sigma_t$ the instantaneous volatility for the Brownian motion component (due to $W_t$) and $J$ is a random variable describing the jump distribution, when a Poisson process $N_t$ experiences a discrete increment.

Note that this model reduces to the industry standard Hull-White model (with time dependent coefficients) if the $JdN_t$ term is omitted.

Calibration

While the mean reversion level $\theta_t$ is often assumed constant, e.g. $\theta_t=\theta$ in conventional interest rate markets, the nature of cryptocurrency markets is such that we need to allow the model to mean revert towards a more recent level. Consequently, we construct an exponentially weighted moving average [EWMA] of daily IPOR values (simple arithmetic mean) to use as the mean reversion level $\theta_t$. We initially used a 30-day span EWMA, to provide a balance between reactivity and stability, though it could be argued that an even shorter span moving average is appropriate based on the period May-July 2021 in Figure 3 below.

Figure 3 - Calibration of θ using exponentially weighted moving average

Figure 3 - Calibration of θ using exponentially weighted moving average

The next step in the calibration is to construct an estimate of the daily volatility $\sigma_t$ for each day, using the standard deviation of intraday increments $\Delta{r_t}$ in the IPOR index, normalised by the square root of the average time increment $\overline{\Delta{t}}$ during the day, to convert into a daily volatility measure.

Once we have this, we can identify candidate jump events by computing a measure of the number of standard deviation move each intraday increment $\Delta{r_t}$ corresponds to

$$ N_{sd}=\frac{\Delta{r_t}}{\sigma_t\sqrt{\overline{\Delta{t}}}} $$

We then select all blocks in the historical time series which are more than a +50 standard deviation move and also an instantaneous increment of more than 300bp in $r_t$ (c.f. the previous) as our signal indicating that a jump event has occurred.