Model Notes
What changed between v1.0 (the book's Appendix C) and v1.1 (the current tools), and what it means for the model's claims.
π The Short Version
v1.0 (book, Appendix C) assumed the debt path was fixed, then computed each bond type's coupons against it. This understated the debt accumulation effect of higher coupons.
v1.1 (current tools) computes a separate debt path for each bond type using its own coupons. This captures the feedback loop the book identifies in Chapter 14: higher coupons β higher debt service β more borrowing β higher future debt service.
The numbers differ, but the direction of every claim still holds β except distress, which is slightly lower in v1.1 than in v1.0. The book's warning about the feedback loop was correct, but the fixed-debt model overstated the magnitude.
π Side-by-Side Comparison
Baseline calibration from config.py. Seed=42. 10,000 paths, 10-year horizon. The book's fixed-debt numbers come from the earlier simulation_engine.py; the current numbers come from run_simulation.py.
| Metric | v1.0 (Book) fixed debt | v1.1 (Current) endogenous debt | Direction |
|---|---|---|---|
| Conventional expected cost | 0.5154 | 0.5154 | Same |
| Productivity expected cost | 0.4994 | 0.4994 | Same |
| Productivity cost saving | β3.1% | β3.1% | Same |
| Conventional distress | 11.05% | 11.05% | Same |
| Productivity distress | 88.17% | 85.68% | Slightly lower (β2.5pp) |
| Inflation-linked cost | 0.4896 | 0.4896 | Same |
| Inflation-linked distress | 70.22% | 67.0% | Lower (β3.2pp) |
| Productivity effective yield | 3.74% | 3.74% | Same |
| Optimal portfolio share | ~30% | 0β45% (regime-dependent) | Sensitive to debt/GDP |
π§ What This Means for the Book's Claims
β Common Objections
No. An ILB protects against inflation. A productivity bond rewards efficiency. If you believe efficiency is the metric we should be rewarding β because it's what we want the economy to produce, not just what we want to protect against β then the productivity bond is the more appropriate instrument. The ILB is a defensive tool; the productivity bond is a strategic one. They serve different purposes and can coexist.
Both are correct for their respective models. The book's number is from the fixed-debt model; the tool's is from the endogenous-debt model. The endogenous model is more realistic and is the current production calibration. The difference is 2.5 percentage points β small enough that the qualitative conclusion (distress rises under variable coupons without a PAM) is unchanged, but large enough that any specific numeric claim in policy discussion should cite which model produced it.
Correct. The simulation tests whether the architecture can work under specified assumptions; it does not prove it will work in reality. The assumptions β about AI shocks, about government capture, about investor behaviour β matter more than the simulation itself. Every one of those assumptions should be challenged before any real-world implementation. The simulation is a tool for designing the instrument, not a forecast of the outcome.
In v1.0, state_generator.py returned a debt path of zeros. The debt stock was passed in as a fixed baseline and used only for computing debt service β it never grew. In v1.1, compute_debt_path() takes a coupon series and rolls forward the debt stock year by year. Each bond type gets its own debt path, so the feedback loop is captured. The GitHub repo (traffictorch/productivity-bond-model) has both models in git history for verification.
π Version History
| Version | Model | Used by | Notes |
|---|---|---|---|
| v1.0 | Fixed debt | Book (2026), Appendix C | Debt stock assumed static per path; coupons applied against it |
| v1.1 | Endogenous debt | Policy Lab, NPI Tracker, current simulation | Debt stock rolled forward per bond type; captures feedback loop |
π Related Pages
- Policy Lab β Interactive scenario builder using v1.1
- Live Simulator β Homepage simulator
- Appendices β Mathematical appendix and full specification
- The Book β Full PDF and audiobook
- Changelog β Version history
- GitHub Repository β Source code and simulation engine
π Last updated: 2026-10-27 Β· Maintained by Traffic Torch