we-solve (beta)

Econ-455-A1 · revision r1 · on Econ-455

GPT 6 Astra Ultra attempt on Econ-455

Exploration · LLM-generated · gpt 6 astra ultra ·

Unverified LLM-generated attempt. Automated screening checks form, honesty and scope, not correctness.

Claim: exploration

Reported by the author as unresolved.

Manuscript

Contents
  1. Recession hysteresis Permanent modes, reversibility, and the cost of capacity repair
  2. Scope and causal interpretation
  3. An exact reversibility criterion
  4. Minimum financing cost and discounted benefits
  5. Finite-horizon identification and its limits
  6. Completion Estimate
  7. References

Recession hysteresis Permanent modes, reversibility, and the cost of capacity repair

Problem ID: E32-451. Status: Unresolved; restricted results and explicit gaps.

Scope and causal interpretation

The target distinguishes a temporary demand contraction from its potentially lasting effects on productive capacity, and asks whether a subsequent, financed intervention reverses them. Yellen's demand-and-supply discussion [yellen] explicitly poses the reversibility question and mentions investment, participation and innovation. The results below give an exact answer in a finite-dimensional capacity-accumulation model. They do not identify its parameters from observed GDP or establish the empirical permanence of any recession.

Let $z_h\in{\mathbb{R}}^n$ contain deviations in log productive capital, labor capacity and technology, possibly for several interacting regions. The capacity index is fixed in advance as $\log K_h=\log K_h^0+w^Tz_h$. The weights $w$ are technological index weights, not fitted output gaps. Spatial spillovers are entries of the matrices below; the unit of causal assignment is the entire connected economy or an independent cluster. After the temporary demand disturbance ends, reset the date to zero and assume its only remaining effect is the state displacement $z_0=v$. Subsequent mean deviations obey \[ z_{h+1}=Az_h+Gu_h,\qquad h\geq0. \tag{1} \] The intervention is a preannounced vector $u=(u_0,\ldots,u_{T-1})$, zero thereafter. It specifies instruments and financing; $G$ includes net capacity effects of both spending and financing. Common additive zero-mean innovations cancel in mean comparisons. Equation (1) is a structural invariance assumption across these interventions, not a regression identity. It is exact within this model; a linear approximation to a nonlinear economy would additionally require an error bound.

Assume $A^h\to P$ in matrix norm. This permits stable recovery modes and unit-root modes, but excludes explosive modes, nontrivial Jordan blocks at one and persistent oscillation. Taking limits in $A^{h+1}=AA^h=A^hA$ gives $AP=PA=P$, and consequently $P^2=P$ by taking limits in $A^hP=P$. No diagonalizability is required.

An exact reversibility criterion

Theorem (Permanent loss and the reachable repair space): Under (1), the no-policy capacity effect and the policy increment are \[ \tau_h=w^TA^hv,\qquad \rho_h=w^T\sum_{s=0}^{\min(h,T)-1}A^{h-1-s}Gu_s. \] Their limits exist and satisfy \[ \tau_\infty=w^TPv,\qquad \rho_\infty=w^TPG\sum_{s=0}^{T-1}u_s. \tag{2} \] For $T\geq1$ and unrestricted signed instruments, the entire permanent state displacement is reversible exactly when $-Pv\in{\operatorname{range}}(PG)$. Reversal of this one capacity index alone requires only that the scalar equation $w^TPG\sum_su_s=-w^TPv$ be solvable. Under componentwise nonnegative instruments the state criterion instead is $-Pv\in\{PGa:a\geq0\}$.
Proof: Induction in (1) gives $z_h=A^hv+\sum_{s<h}A^{h-1-s}Gu_s$, with terms $s\geq T$ absent. Subtracting the untreated state establishes the finite effects. For $h\geq T$, only $T$ terms occur, each converging to $PGu_s$, proving (2). The post-intervention state limit is $Pv+PG\sum_su_s$. The set of sums of $T$ signed controls is ${\mathbb{R}}^m$, since the first control can equal any desired sum and the others can be zero. With nonnegative controls that set is ${\mathbb{R}}_+^m$. Setting the limit or its $w$-projection to zero gives the stated equivalences. ∎
If all eigenmodes recover so that $P=0$, finite persistence is compatible with zero permanent effect. Conversely, a permanent technology or participation loss cannot be repaired by an instrument loading only on stable capital modes. Restoring a weighted index can conceal opposing persistent component losses; the two criteria deliberately differ.

Minimum financing cost and discounted benefits

Stack controls into $U\in{\mathbb{R}}^{mT}$ and put $C=(PG,\ldots,PG)$. Let $R$ be a symmetric positive definite cost matrix; $U^TRU/2$ is a declared real-resource and financing-distortion cost in a common numeraire, including timing weights. Transfers alone are not costs.
Theorem (Least-cost permanent repair): If $b=-Pv$ lies in ${\operatorname{range}}(C)$, the unique minimum of $U^TRU/2$ subject to $CU=b$ is \[ U^*=R^{-1}C^T(CR^{-1}C^T)^+b,\qquad J^*=\tfrac12 b^T(CR^{-1}C^T)^+b, \tag{3} \] where $^+$ is the Moore--Penrose inverse. If $b$ is outside that range, repair is infeasible. Formula (3) is for signed controls; sign and budget limits require the corresponding constrained quadratic program.
Proof: Set $y=R^{1/2}U$ and $D=CR^{-1/2}$. The problem minimizes $\|y\|^2/2$ subject to $Dy=b$. In an orthogonal singular-value decomposition $D=V\Sigma W^T$, feasibility sets each coordinate of $W^Ty$ associated with a positive singular value equal to the corresponding coordinate of $V^Tb$ divided by that value; zero singular values require the corresponding coordinate of $V^Tb$ to vanish. The remaining free coordinates uniquely minimize the squared norm at zero. Thus $y^*=D^T(DD^T)^+b$, with squared norm $b^T(DD^T)^+b$. Substitution yields (3) and the infeasibility statement. ∎

For $\delta\in(0,1)$, let $\eta$ be the declared marginal social value of a unit of this log-capacity index. This is a linear welfare criterion, not consumption utility inferred from the index. Since $A^h$ converges, its powers are bounded and $\sum_h\delta^hA^h=(I-\delta A)^{-1}$. Hence the discounted capacity gain from a finite control is exactly \[ \sum_{h\geq0}\delta^h\eta\rho_h =\eta\sum_{s=0}^{T-1}\delta^{s+1} w^T(I-\delta A)^{-1}Gu_s. \tag{4} \] Indeed absolute convergence permits switching the finite control sum with the infinite horizon sum, and the geometric series follows by telescoping its finite partial sums; the remainder $\delta^{H+1}A^{H+1}$ tends to zero. Subtracting the actual $U^TRU/2$ gives net welfare in this declared criterion. Least-cost permanent repair need not maximize (4), because transitory effects and the timing of repair also have value.

Finite-horizon identification and its limits

Fix baseline covariates $X$, contraction $D\in\{0,1\}$, an intermediate history $H$ measured before policy, and policy $A_0\in\{0,1\}$. Here $A_0$ is a treatment indicator, not the transition matrix $A$. Let $Y_h(d,a)=\log K_h(d,a)$ and $H(d)$ be well-defined cluster potential outcomes. Assume consistency, positive conditional treatment probabilities, integrable outcomes, and sequential exchangeability: \[ (H(d),Y_h(d,a))\perp D\mid X,\qquad Y_h(d,a)\perp A_0\mid X,D=d,H. \] The second condition is understood within the $D=d$ potential history; it excludes residual policy selection after conditioning on that history.
Proposition (The required sequential adjustment): Under these conditions the regime mean is \[ m_h(d,a)=\int\!\int {\mathbb{E}}[Y_h\mid X=x,D=d,H=h',A_0=a]\, dP(h'\mid x,D=d)\,dP_X(x). \tag{5} \] Therefore $\tau_h=m_h(1,0)-m_h(0,0)$ and $\rho_h=m_h(1,1)-m_h(1,0)$ are identified for observed horizons.
Proof: Condition first on $X$ and $H(d)$. Sequential exchangeability and consistency identify the conditional outcome mean by the observed conditional mean in (5). Baseline exchangeability and consistency identify the law of $H(d)$ given $X$ by $P(H\mid X,D=d)$. Iterated expectation then gives (5). Differences of marginal means do not require a joint cross-regime distribution. ∎
Conditioning on the same realized post-contraction history in both contraction regimes would change the target. Migration must be handled through a fixed population or a stated exposure mapping, and measurement revisions must be harmonized before interpreting $Y_h$ as capacity. Component decompositions of $w^Tz_h$ are index accounting, not identified causal mediation effects.
Proposition (Exact finite observational equivalence): For every finite $H\geq1$, an unrestricted post-shock accumulation class contains two deterministic economies with identical regime outcomes through $H$, a temporary initial demand shock, and different permanent effects. This remains true even if their subsequent state transitions are linear and time invariant, when state dimension is unrestricted.
Proof: The permanent economy has scalar state $z_0=-1$, transition $z_{h+1}=z_h$ and log-capacity deviation $z_h$. In the recovering economy use $H+1$ states, all initially $-1$, transition $(z_1,\ldots,z_{H+1})\mapsto(z_2,\ldots,z_{H+1},0)$, and observe the first state. The initial shock acts only at date zero in both economies. The observed deviations equal $-1$ at dates $0,\ldots,H$ in both, but the recovering deviation is zero thereafter whereas the permanent one remains $-1$. Give policy zero capacity effect in both and set $K_h=\exp(Y_h)$, so all joint regime data coincide through $H$ and capacity is positive. Both transitions are time invariant; one has a unit mode and the other is nilpotent. Their permanent contraction effects are respectively $-1$ and zero. ∎
This construction does not defeat identification in a known, fully observed, low-dimensional model. It pinpoints the extra content required for extrapolation: state coverage, dimension restrictions, invariance and information about the permanent projection $P$. Estimating these with credible demand and policy variation, validating financing effects, and allowing irreversible nonlinear exit remain unresolved parts of the target.

Completion Estimate

60%

This is a post hoc, heuristic estimate for the full catalogue target. The linear state model yields exact permanent damage and repair conditions, least cost intervention formulas, and a finite horizon observational equivalence showing why permanence is not generally identified. Credible demand and stimulus variation, financing effects, nonlinear exit mechanisms, and an empirically identified welfare model are still missing.

References