# Step 286: investment response on matched country-years

Step 282 compared total fixed investment across 38 OECD members with private fixed investment in seven members. This audit separates the change in sample from the change in investment measure, using the same archived World Bank WDI panel. Run `python scripts/investment_coverage_v286.py` from the repository root to reproduce the estimates and simulation counts.

## Investment and unemployment

The outcome is the annual change in investment as a percentage of GDP. The reported slope adds the current and one-year lag coefficients on changes in unemployment, controlling for current and lagged real GDP growth. The panel uses country and year effects and clustered standard errors by country. The intercept is explicit here because the v282 helper drops a reference category from both dummy sets. The full-panel point estimate remains −0.4343 at four decimal places. The Israel-only regressions use an intercept and Newey–West errors with two lags.

| Sample and outcome | Observations | Slope | Approximate 95% interval | Implied model decay |
|---|---:|---:|---:|---:|
| All 38 countries, total investment | 1,206 | −0.434 | [−0.638, −0.231] | 1.27 (v282 anchor) |
| Matched seven countries, total investment | 194 | −0.249 | [−0.484, −0.014] | 0.73 |
| Same 194 observations, private investment | 194 | −0.206 | [−0.414, 0.001] | 0.61 |
| Same 194 observations, private minus total | 194 | 0.042 | [−0.047, 0.132] | |
| All countries, GDP did not contract in either year | 917 | −0.353 | [−0.538, −0.167] | 1.04 |
| All countries, GDP contracted in at least one of the two years | 289 | −0.474 | [−0.669, −0.279] | 1.39 |
| Israel, 1993–2024 | 32 | −0.668 | [−1.285, −0.051] | |
| Israel, 2000–2024 | 25 | −0.262 | [−0.918, 0.394] | |

The matched sample covers Australia, Chile, Costa Rica, Korea, Mexico, Poland and Türkiye. Restricting the total-investment sample accounts for about four-fifths of the numerical change between the two original slopes. The within-sample difference is imprecise. Inference with seven clusters is fragile and the intervals use the v282 normal approximation. The two GDP regimes are descriptive subsamples selected on contemporaneous and lagged growth. The table provides no causal estimate of automation-induced displacement and no formal test of a difference between the GDP regimes. Private fixed investment still includes residential investment. The Israel-only estimate since 2000 is imprecise and uses a different specification from the earlier panel interaction.

## Consumption and debt screen

Every entry counts cases out of 432. Both displaced groups must maintain at least 90% of initial real nonhousing consumption in every year, and peak total net public debt must stay at or below 90% of output. Both levy-funded policies share the v285 ceiling of 15% of positive profits across the benefit and demand support levies, with priority for the benefit. Support is limited to 5% of output. The response to the policy rate varies between 0.60 and 0.15.

| Investment decay | No support (0.60 / 0.15) | Support at any demand gap (0.60 / 0.15) | Support only at zero rate (0.60 / 0.15) |
|---:|---:|---:|---:|
| 0.61 | 266 / 251 | 276 / 278 | 266 / 267 |
| 0.73 | 252 / 208 | 268 / 269 | 260 / 233 |
| 1.04 | 135 / 79 | 205 / 166 | 172 / 142 |
| 1.27 | 60 / 27 | 134 / 91 | 119 / 81 |
| 1.39 | 32 / 18 | 89 / 63 | 80 / 52 |

Support at any demand gap increases joint passes in all ten comparisons. That ranking is specific to the joint screen. For example, at decay 1.04 and rate sensitivity 0.60, support only at zero preserves consumption in 246 cases versus 211 with support at any gap. The joint ranking reverses once the debt screen is applied. At decay 1.27 both rate sensitivities reproduce the previous shared-ceiling results. The large change in pass counts across the decay settings warrants reporting the range and keeping 1.27 as a stress benchmark. The grid is illustrative and its shares have no probabilistic interpretation.