get_r_gov clips the sovereign interest rate at zero:
r_gov = np.maximum(
p.r_gov_scale[t] * r - p.r_gov_shift[t]
+ p.r_gov_DY * DY_ratio + p.r_gov_DY2 * DY_ratio**2,
0.00,
)There is no way to switch it off, so a country whose government genuinely pays a negative real rate cannot be represented.
Japan is that country. General government net interest is about 0.43% of GDP on net debt of 114.8%, roughly 0.375% nominal, which against realised inflation is −0.6% real. Calibrating the wedge to that works — it returns −0.6023% — and the floor then discards it and reports 0.0000.
Why it matters. The debt-stabilising primary balance is pb* = (r_gov − g)/(1 + g) · D/Y. Clipping the rate from −0.60% to 0.00% raises the primary balance the country must run by 0.68 percentage points of GDP. For Japan that is the difference between a model whose fiscal stance matches the actual accounts and one that demands a primary surplus Japan has never run. On the transition path the clipped version has debt rising where Japan's is falling; with the true rate it tracks the published projection to within half a percentage point over three years.
Negative values are safe. r_gov enters the model linearly everywhere — no division by it, no powers, no logs. It has two consumers:
debt_service = r_gov * D # fiscal.get_D_ss / get_D_t
r_p = (r_gov*D + r_K*K) / (D + K) # aggregates.get_r_p
Both are linear and the r_p denominator is strictly positive. A negative r_gov simply means the government is a net receiver on its debt position, and the household portfolio return blends a negative debt leg with a positive capital leg. I have been running Japan this way against v0.19.0 with a local patch: the steady state solves, the transition converges, and the resource constraint holds in 318 of 320 periods.
Suggested fix. Make the floor a parameter rather than a constant — r_gov_floor, default 0.0. Existing calibrations are unaffected, since np.maximum(x, 0.0) is exactly what happens today, and a country with a negative sovereign rate can lower it. Happy to open a PR.
cc @jdebacker @rickecon
get_r_govclips the sovereign interest rate at zero:There is no way to switch it off, so a country whose government genuinely pays a negative real rate cannot be represented.
Japan is that country. General government net interest is about 0.43% of GDP on net debt of 114.8%, roughly 0.375% nominal, which against realised inflation is −0.6% real. Calibrating the wedge to that works — it returns −0.6023% — and the floor then discards it and reports
0.0000.Why it matters. The debt-stabilising primary balance is
pb* = (r_gov − g)/(1 + g) · D/Y. Clipping the rate from −0.60% to 0.00% raises the primary balance the country must run by 0.68 percentage points of GDP. For Japan that is the difference between a model whose fiscal stance matches the actual accounts and one that demands a primary surplus Japan has never run. On the transition path the clipped version has debt rising where Japan's is falling; with the true rate it tracks the published projection to within half a percentage point over three years.Negative values are safe.
r_goventers the model linearly everywhere — no division by it, no powers, no logs. It has two consumers:Both are linear and the
r_pdenominator is strictly positive. A negativer_govsimply means the government is a net receiver on its debt position, and the household portfolio return blends a negative debt leg with a positive capital leg. I have been running Japan this way against v0.19.0 with a local patch: the steady state solves, the transition converges, and the resource constraint holds in 318 of 320 periods.Suggested fix. Make the floor a parameter rather than a constant —
r_gov_floor, default0.0. Existing calibrations are unaffected, sincenp.maximum(x, 0.0)is exactly what happens today, and a country with a negative sovereign rate can lower it. Happy to open a PR.cc @jdebacker @rickecon