Four Decades of High Yield Default Rates: A Forecast
September 25, 2015
A bear market in treasuries means a 250 bps rise in spread levels on existing high yield bonds. These bonds could easily price in the mid-80s.
Forecasted default rates on high yield debt rise to 5%, not accounting for spike in default rates due to regime-switching to a high-correlation, stressed market. A rise to 12% shouldn’t be discounted out of hand given the oil price shock.
Forecasted hypothetical returns given combined default and spread levels are -3.5%. Something worse could be looming.
High yield bonds have been around a lot longer than Michael Milken and Blondie’s “heart of Glass”. They just didn’t have a credit rating. That is to say: they predate the rating agencies that attach variations of BB, B, and C grades on a bond. I also suspect that the earliest high yield bond investors got savagely burned by them. The losses took decades of healing—perhaps even a generation—before anyone had the stomach for them. The historical data below makes this point.
So the high yield market may not have made it without the bull market in treasuries enjoyed over the last several decades. Without its protective influence, it never would have had a chance to really thrive. It would be a maligned shadow of its present self. My Uncle Larry was always suspicious of it.
My point is this: favorable capital market rates anchored by the treasury yield curve and very liquid duration-matched hedging capabilities made the explosive growth in high yield issuance possible. It is likely so big now that no longer needs this support. True or not, a treasury bear market implies the loss of the favorable pricing anchor and liquid hedging possibilities will create serious dislocations in what is now a trillion dollar high yield market.
So is a treasury bear market likely? No. But it isn’t impossible either. It is very possible that interest rates will remain at current levels even after wage pressures build, because the Fed finds itself incapable of entering a rising rate environment without causing major market disruptions. So it will hold off until major market disruptions occur, and then act. This is a tail risk to take seriously whether you are a credit guy or a bond dude.
For this reason, I took four decades of default rate data and put together a forecast.
Conceptually, it is a little off-target to talk about a tail risk because it implies a standard empirical distribution that holds up robustly across all times and places. There is a uniformity about bonds and an independence about their default characteristics. The reality is that bond defaults are not independent of each other; they are correlated. Nor is one bond the same as every other in the characteristics that matter the most. So the phenomena of defaults are best explained in a bi-modal way. Behavior is driven by drastically different states of nature, or regimes. And these regimes switch back and forth from “stress” to “normalcy” quite a lot.
In terms of method, all quantitative models depend on some form of mean-reversion where assets oscillate predictably around some long-term average. They won’t work otherwise, because this implies the existence of some profitable investment strategies that trade based on convergence to that average.
I would not blame you for being suspicious on this point. Kay Giesecke looked at 150 years of corporate defaults and found value-weighted defaults rates in the days around the Civil War looking like something from another planet. War can do that to debt. But independent of a war aftermath, default rates were high even outside of periods of distress where they spiked. See chart to the right (from http://www.nber.org/papers/w15848.pdf).
But even during the darkest days, default rates still resemble something like the 10%-12% or so that we see under the sign of the great treasury bull market of our lifetimes (at the peak it was 16%). This constancy speaks to unavoidable idiosyncratic drivers of default that prevail no matter the financial ecosystem conditions. One is operating leverage (debt to EBITDA). When a company is operating with a leverage ratio greater than 6, it means real trouble for viability. The second is free cash flow coverage (EBITDA – capex) to debt). The smaller this number is, the less likely the debt pile is serviceable. No amount of interest rate favorability can overcome crippling problems that rule the laws of default dynamics.
Further, there appears to be little more mean-reverting than a credit spread. So I take the view that it is at least reasonable to model high-yield default rates based on their possessing some brand of the stationary property. Stationarity likely only holds only in a local sense, but the auspices of our bull-market in treasuries create the local conditions we need to consider. This is a model risk we wish to take on, as we can build an uncertainty factor to account for deterioration of local conditions.
The forecast values are generated using the Winters method of updating equations similar to exponential smoothing to fit parameters for the model
What you see below is a head-to-head comparison of realized defaults (purple) and model-implied (light blue) values. The model consistently underestimates default levels, but it updates from the time series to take spike in realized defaults seriously.
The forecast prediction value t periods ahead is the solution to the equation
This provides a base case default rate, shown as a continuation of the light blue model values. Because of the underestimation problem, an “adverse case” is built by trend adjusting a 3-sigma move in defaults (orange). This gives us a view of what a “spike” would look like in any given year of the forecast, allowing trend to let it rise, consistent with greater uncertainty the further out in the future you go.
Model implied default rates imply a 200 basis point rise in yields. Worst case default rates exceed 10% but do not exceed 12%.
This really doesn’t come close to a “worst case” scenario. Even in sample the worst case is less than the highest realized default rate. What the forecast represents is a reasonable view of what future default rates will look like on average, and a very high-probability view of how bad it could get, based on the last 4 decades of regime switching.
The model: Bonds are discrete “particles” of varied size and character which diffuse and interact with each other and committing capital to instruments that will ultimately mature or default. Their growth and survival is subject to the fate of the capital instruments to which they commit capital. This forecast accountings for regime switching from correlated stress conditions to more uncorrelated benign conditions by providing a default rate forecast under stress and also under more normal conditions.
A 5Y Treasury yielding 150 bps is not normal. The implications: as of September, the BofA US High-Yield Index has a weighted duration of 4.2 years and a current yield of 7.14%. The yield on the closest duration matched UST (the 5Y) is 1.55%. A base case default rate of 5% will imply a 250 bps spread change, -9% losses from spread change, and a further -2.7% from defaults. At current yield, the combined upick in spreads and default losses implies return in the -3.5% area. This means coupons need to rise on new issues and existing debt could price in the 80s.
The residual: A bull market in treasuries grants this population a favorable habitat (call it an oasis) as opposed to a much more risky environment (call it a desert). It is not clear how much these ecosystem conditions matter, but they do alter fundamentals. Based on the explosive growth in high yield debt in the last four decades coinciding with the T-bull, I would venture the favorable ecosystem state is crucial to the high-yield market. Not to stretch the analogy too much, but it seems likely that the immediate market adaption to a liquidity desert will be to downsize. The survivors will either get even cheaper that the implied base case and a lot will evolve into rising stars (investment grade debt).
This natural change plus a switch to stress-level defaults due to a higher spike in defaults due to the oil shock causing non-viability… well, that implies carnage.