מסמך טכני מיומן המחקר · באנגלית, כפי שנכתב בשלב העבודה
Step 248: borrowing closures instead of stop rules
Why. Every count in the grid is conditional on completion, and completion is set by two stop rules: the government cannot borrow (the path stops if existing spending and interest exceed revenue), and firms stop when their cash turns negative (investment is the residual of output less government purchases and consumption, profits are paid out in full). Of 576 cases only 202 complete. A referee would ask whether the results describe the economy or the stop rules.
What. scripts/closure_model_v248.py is a copy of the v98 household model (archived file untouched) with two closures:
- Government: may issue bonds while net debt stays below a cap (90% or 120% of output; the base
year is 60%). Benefits are paid in full within that room and rationed beyond it. Existing spending and interest are always paid; new bonds are bought by households pro rata to their cash, and interest and redemptions go to holders pro rata to holdings.
- Firms: may run a bank overdraft (bank money owned by the capital owners) at 3%.
With the strict settings the new model reproduces v98 exactly: 202 completed, 75 preserved, 42 matched controls (tested). Generated by scripts/closures_v248.py; archived in data/processed/closures_v248.json. Benefit indexed to goods prices, no-credit variant, capital feedback 0.3, as in the grid.
All 576 cases
| Closure | Completed | Both displaced kept 90% | Share of all 576 | Matched controls |
|---|---|---|---|---|
| Strict (v98) | 202 | 75 | 13% | 42 |
| Firm overdraft only | 432 | 264 | 46% | 288 |
| Government cap 90% only | 346 | 112 | 19% | 258 |
| Both, cap 90% | 576 | 301 | 52% | 576 |
| Both, cap 120% | 576 | 335 | 58% | 576 |
Most stops in the grid are firm liquidity stops (230 of 374). With both closures every path completes. In 144 cases existing spending and interest alone exceed revenue, so net debt rises past the cap (to at most 1.8 times output at a 90% cap): the stop has been turned into a debt build-up, which is the honest description of those cases.
Which patterns survive (both closures, cap 90%, share of all 576 preserved)
| Factor | Values | Preserved |
|---|---|---|
| Pass-through | 0 / 0.5 / 1 | 60% / 57% / 40% |
| Share of profits within reach of tax | 0.5 / 1 | 35% / 69% |
| Profit-tax rate | 25% / 35% | 40% / 65% |
| Capital ownership | concentrated / broad | 42% / 63% |
| Tenure | renter / outright / fixed mortgage / reset mortgage | 54% / 63% / 56% / 37% |
| Productivity gain | 0.25 / 0.5 / 1 | 23% / 54% / 79% |
- Survive: full pass-through lowers preservation; tax reach and the tax rate are the strongest
factors; broad ownership helps; rate-reset mortgage holders are hit hardest.
- Reverses: productivity. In the grid no case with doubled productivity preserved the
displaced (0 of 24). With the closures 79% do. The grid's result was selection by the stop rules, as the results chapter already suspected; with the closures, more productivity helps.
- Weakens: the renter penalty (54% against 63% for outright owners). Step 247 shows it is
mainly a deflation effect.
Relative position
With both closures every case has a matched control. Both displaced cohorts end below the control in 90% of the 576 pairs (median -24% without consumer debt, -20% with), and retained workers end above it in 96% (median +31%). The finding survives, but as "almost always", not "in all cases".
What changes in the text
- Headline counts should be reported over all cases with the closures, with the strict grid as a
sensitivity: "the displaced keep their consumption in about half of the cases (52%-58%)".
- The productivity paragraph should say that with borrowing closures more productivity helps.
- "In all 42 comparisons" becomes "in 90% of 576 comparisons".
Limits
- The interest rate on public debt stays at 3% whatever the debt level; there is no risk premium,
so the debt cap is the only fiscal limit.
- The overdraft has no limit and no default; it measures how much finance the residual-investment
rule would need, not whether a bank would lend it.
- The caps (90%, 120%) are research choices, not estimated limits.