This whitepaper includes a short summary of Moment’s pricing methodology for corporate bonds, as well as two analyses that benchmark the performance of Moment’s fair value against publicly-reported TRACE transaction data and ETF-reported vendor marks.

Pricing Methodology

Moment follows a waterfall approach for corporate bond pricing, relying on direct observations (e.g. TRACE trades, dealer quotes) for sufficiently liquid bonds and falling back to a factor-based model for highly illiquid bonds without direct observations.

Direct observation modeling

Each day, Moment receives TRACE data on roughly 10,000 unique corporate bonds, as well as dealer quote data on nearly 40,000 unique corporate bonds. These data cover nearly the entire universe of corporate bonds, allowing Moment to price most bonds using direct observation-based models. Moment offers several direct observation-based pricing models, but they each follow the same high-level procedure for evaluating a bond’s fair value:

  1. Adjust historical trade and quote spreads to the present. Because most bonds are traded and quoted infrequently, there can be significant displacements in interest rates and credit markets between the time of the observation and the time of evaluation. To account for this, Moment uses a proprietary factor model to “beta-adjust” historical observations to the present.
  2. Estimate the distribution of each observation around the theoretical fair value. After adjusting historical observations to the present, Moment estimates the mean and variance of the distribution of the observation around the theoretical fair value. The estimated mean and variance are functions of several factors, including the type of observation (trade vs quote), the counterparties to the observation (dealer-to-dealer trade vs dealer-to-customer trade), and the staleness of the observation.
  3. Fit a robust likelihood model to the data to determine a spread. Once each observation has an estimated distribution around the theoretical fair value, estimating a fair value spread simply becomes a maximum likelihood problem. Because Moment uses fat-tailed distributions to model observations, Moment’s fair value model natively handles outlier rejection.

Inferred factor modeling

While Moment receives direct observations for most corporate bonds, there are still some bonds that Moment’s data sources do not cover. When this is the case, Moment’s fair value model falls back to a factor-based model that uses observations in related bonds to estimate a spread for a very illiquid bond. Specifically, Moment’s factor-based model constructs an issuer curve and, if no data is available for the issuer, a sector curve that estimates the relationship between a bond’s duration and yield within the issuer and sector group.

Untitled

Intraday TRACE performance

The sections below outline Moment’s methodology for assessing intraday model performance, as well as provide historical results for the period from Jan 2022 to Feb 2023.

Inclusion criteria

Moment measures the performance of its pricing models using publicly-reported TRACE data. To be included in the performance analysis, a TRACE trade must meet the following conditions:

  1. The trade must occur during normal market hours
  2. The trade quantity must be at least 100,000 (100 bonds)
  3. Both counterparties to the trade must be dealers

The rationale for the last condition is that dealer-to-customer trades include some degree of bid-offer; dealer-to-dealer trades are assumed to occur at mid. Canceled and clearly erroneous trades are also excluded from the analysis.

Performance metrics