Scaling quantitative bond analysis: How difficult is it?

Bloomberg Professional Services

Fixed income is a naturally quantitative asset class: the investor claims a predetermined, and thus quantifiable, stream of cash flows. This implies that greater accessibility of data and processing tools would make it easier for market participants to scale up their competitive advantage for it to be applied systematically.

Except, the challenge resides with core bond mechanics. The variety of cash flow structures available amongst a vast number of securities create an uneven playing field. Additionally, as the relationship between a price and its corresponding yield is nonlinear or convex, each instrument is affected differently by changes in interest rates or credit spreads, depending on its cash flow pattern and pricing level.

Unlike with any other asset class, bond analysis focuses on yields and spreads rather than on prices. At the same time, mark-to-market returns are directly linked to price moves. Therefore, a dynamic strategy needs to anticipate shifts in rates or spreads, as well as calibrate resources based on the impact those moves are expected to have on prices.

Four normalization pillars

The consequence of working with unlevel, and at times contrasting inputs, is that quantitative analysis is hard to scale. While no universal method is known to even out the field, four normalization pillars can be successfully combined: pattern, pricing, predictability, and performance.

• Pattern

A variety of cash flow patterns – including long- or short-dated instruments, options, amortization, fixed or floating coupons – cater for a wide range of preferences. While investors’ choices are driven by different motivations such as return targets or asset-liability constraints, a key determinant is the price sensitivity of those structures to changes in rates or spreads. An approach that brings securities with dissimilar cash flow patterns to a meaningful common denominator is essential.

For example, the following three bonds from the same issuer have similar cash flow structures: MSFT 4.2 11/03/35, MSFT 3.45 08/08/36, MSFT 4.1 02/06/37. It is undetermined which two instruments have the most closely matching structures, as 2036 and 2037 have a more similar distribution of cash flows based on their durations, whereas 2035 and 2036 display a closer sensitivity to changes in rates and spreads as of the time of this writing.

• Pricing

Markets tend to watch rates and spreads interchangeably with varying emphasis, as simultaneously considering both can be challenging and occasionally resurface contrasts. As a result, market participants specialize in either credit or rates and focus their analysis on the component underlying their bets.

For example, a participant in the credit market would predominantly analyze spreads on the basis that the rate component is to be hedged out. Conversely, rates markets would naturally ignore credit spreads. This means that the tensions between the two elements are more likely to be overlooked than taken advantage of. Developing a method that consistently balances out both parameters can therefore unlock opportunities.

• Predictability

Pricing predictability is a cornerstone for strategies applied in the context of multiple instruments, e.g. portfolios. Traditional funds tend to prefer a lower correlation between asset prices to diversify the effect of market movements, whereas market-neutral strategies are likely to favor synchronized behaviors. Furthermore, differentiating between systematic and idiosyncratic factors ensures the investment decision is articulated with further intention.

• Performance

Fully ascertaining historical bond performance across a large set of securities is possibly the most difficult to achieve. Namely, the price of a bond moving higher than another’s doesn’t necessarily mean one has outperformed the other. Conversely, the magnitude of yield and spread shifts might not be representative of their significance.

A popular metric for normalizing shifts for a single security based on their standard deviation, the Z-Score, is bound by the same limitations. For example, the spreads of two bonds with durations of 5 and 25, move 10 basis points tighter from their own average with a standard deviation of 5 bps over the same time frame. Viewed in isolation, those shifts and their resulting Z-Scores of 2 appear to be similar.

However, adding the dimension of capital allocated to either bond to monetize such shift demonstrates that a position of $1,000,000 in each of the two bonds would have resulted in a $5,000 gain for the former resp. $25,000 for the latter. This comparison however is not equitable as the risk associated with the higher duration security is substantially higher.

Introducing a third security to benchmark the performance of the initial two bonds against would level the playing field for this comparison. The benchmark security has a duration of 10 and its spread stayed unchanged over the analysis time frame. A total of $2,000,000 is allocated towards each of the two pairs consisting of the benchmark bond and one of the other two bonds.

The first pair including the bond with a lower duration would have recorded mark-to-market gains of approximately $6,667, whereas the second pair would have made $14,286. Alternatively, a 15 bp shift in the former security or a 7 bp shift in the latter would have resulted in the same gain of $10,000. Those breakeven levels are however relative to the security used as a benchmark and they are bound to change when bonds are paired with a security with a higher or lower duration.

This reveals that the question of bond performance can only be answered with precision in a well-defined context based on like-for-like risk and capital constraints. Working out an equitable distribution of risk and capital simultaneously is nevertheless a challenging task requiring an effective approach to minimize the computational load.

Performing a large-scale bond analysis

While bond analysis is an inherently quantitative process, scaling it is hard. Leveling out the field for heterogeneous inputs across four dimensions – cash-flow patterns and current pricing, as well as the predictability and performance of historical pricing – can facilitate scaling up one’s competitive edge.

At Bloomberg, we build on our data and quantitative analytics capabilities to forge a foundation for clients performing large-scale research. From our extensive reference and pricing data and through our established financial analytics, we normalize and synthesize the relationships between bonds so that they can be explored systematically.