Discussion about this post

User's avatar
Ronald Cook's avatar

I find little fault with your logic, but your four requirements are all demand-side, single-generating-process metrics. They answer: "Given historical bond market dynamics, is the compensation adequate for a long-term saver?" Every requirement assumes the bond market's return distribution is drawn from one population; the historical one where term premium, real rates, breakevens, and convexity are the sufficient statistics.

My point (expanded on below) is that the current bond market has at least four distinct generating processes operating simultaneously, and your framework collapses them into one:

Process 1 — Classical risk premium (your model). The term premium normalizes, real rates are positive, and bonds compensate for duration. This is the process your four requirements measure. It says: buy.

Process 2 — Reserve/trust erosion. ROW central banks and sovereign wealth funds pulling back from Treasuries not because of yield inadequacy but because of counterparty mistrust of the US fiscal and institutional trajectory. This generating process has nothing to do with term premium or real rates. It operates on a completely different parameter space: geopolitical alignment, sanctions risk, dollar weaponization history, institutional credibility. Your framework literally cannot see this process because none of your four requirements capture it. A bond can meet all four requirements and still get sold by a reserve manager who no longer trusts the issuer.

Process 3 — Fiscal dominance. Accelerating deficits with no mechanism for correction. No adults in the room. The supply of Treasuries is growing faster than any demand-side analysis can absorb. You list "government supply" as one of your 10 short-term bearish factors, but you treat it as cyclical; a short-term headwind within a structurally sound long-term asset. Fiscal dominance says it's not cyclical. It's a regime change in the generating process itself. The supply trajectory is not mean-reverting. It's accelerating. And every additional dollar of issuance changes the return distribution of every existing bond in ways that the historical backtests (1973-present) never encountered at this scale.

Process 4 — Energy/geopolitical supply shock. Trump/Hegseth policy driving energy costs structurally higher. This attacks your weakest requirement (inflation compensation) from a direction your framework treats as exogenous noise rather than a persistent generating process. If oil isn't cycling; if it's being structurally repriced by policy incompetence, then the "meh but not awful" assessment on inflation comp isn't just the weak link. It's the link that's actively corroding.

The Homogeneous Population Fallacy diagnosis:

When you demix these four processes, the picture is radically different:

Process DS Framework Sees It? Direction for Bonds Persistence

1. Classical risk premium YES — this is his entire model Bullish Mean-reverting (his backtest works)

2. Reserve/trust erosion NO — outside his parameter space Bearish Structural / accelerating

3. Fiscal dominance Partially — listed as short-term factor Bearish Structural / accelerating

4. Energy policy shock Partially — via inflation comp caveat Bearish Policy-dependent, not cyclical

You are modeling Process 1, acknowledging Processes 3 and 4 as short-term headwinds, and seem to be ignoring Process 2. Your framework is adequate if and only if Process 1 dominates the mixture. We may be in a regime where Processes 2-4 collectively outweigh Process 1, and they've never co-occurred at this intensity in your backtest window.

The 1973-present backtest is the tell. Your strongest empirical evidence, "unlevered long bonds nearly matched SPX at 60% of the risk" is drawn from a period when the US was the unquestioned reserve issuer, deficits were intermittently controlled, energy shocks were temporary, and ROW had no alternative to Treasuries. Every single one of those conditions is now degraded or reversing. The backtest is a single-population sample from a regime that may no longer be the generating process.

Ronald Cook's avatar

Whatever. I was just trying to engage. I have already added RFIX to my own portfolio as a high-convexity bet on bonds, since the PFIX/RFIX ratios have been compressing and other metrics (Diado first order coupling metrics) suggest the switchover to the bond side might be sharp rather than gradual

14 more comments...

Ready for more?