we-solve (beta)

Econ-376 · economics · Economics

Synthetic-control inference with few treated units

LLM status Open unverified

Statement

Original identifier: E6 (atlas item 106). Field: Econometrics and causal inference. Classification: editorial research family with established restricted solutions; current open status not certified.

Definitions and assumptions

Observe a panel $Y_{it}$ for units $i=1,\ldots,N+1$ and times $t=1,\ldots,T$, with unit $1$ treated only after $T_0<T$ and other units untreated. Let $Y_{it}(d)$ be outcomes under a specified treatment regime $d\in\{0,1\}$, with no anticipation and no spillovers to donors. A candidate untreated-outcome model is $Y_{it}(0)=a_i+b_t+\lambda_i^\top f_t+u_{it}$, where $\lambda_i,f_t\in\mathbb R^r$, $r$ is fixed or has stated growth, and $u_{it}$ follows a specified time-series law. The realized post-treatment contrast is $\theta=(T-T_0)^{-1}\sum_{t>T_0}[Y_{1t}(1)-Y_{1t}(0)]$. Observed outcomes equal $Y_{1t}(1)$ for $t>T_0$ and $Y_{it}(0)$ otherwise.

Formal research question

For a declared stochastic-panel or random-assignment experiment, construct intervals for $\theta$ that include donor selection, counterfactual-model ambiguity and post-treatment prediction noise. Let $\mathcal P_{N,T}$ restrict factor strength, temporal dependence, donor contamination and an approximation error $\Delta_{N,T}$ for representing unit $1$ by donor weights $w_j\ge0$, $\sum_{j=2}^{N+1}w_j=1$. Seek \[ \liminf_{N,T\to\infty}\inf_{P\in\mathcal P_{N,T}} \Pr_P\{\theta\in C_{N,T}\}\ge1-\alpha,\qquad\alpha\in(0,1), \] with explicit relative growth of $T_0$, $T-T_0$ and $N$. Determine when diameter decreases, and lower bounds when an unspanned factor or a single unpredictable post-treatment shock cannot be learned. Also give finite-sample randomization guarantees if the treated unit was actually assigned by a known design.

Interpretation and scope

The coverage probability must be tied to a real experiment; with a fixed treated country there is no automatic exchangeability of its treatment label with donor labels. Placebo ranks may be diagnostics without being exact randomization $p$-values. Good pretreatment fit does not prove that latent factor relationships persist after treatment. Inference for a conditional mean effect differs from predicting a realized untreated trajectory, which includes irreducible idiosyncratic noise. A growing donor pool cannot necessarily compensate for a fixed post-treatment horizon. Negative donor weights, extrapolation and matrix completion require distinct approximation assumptions.

Source audit and status

[S1] and [S2] give established synthetic-control constructions and applications; [S3] develops synthetic difference-in-differences. Their assumptions and asymptotic regimes should not be converted into a universal one-treated-unit guarantee. The residual research question is scoped robustness to model choice and limited post-treatment information. This is an editorial formulation supported by the literature, with no certified claim that all its variants remain open in 2026.

References

    },leftmargin=*]

  1. Alberto Abadie, Alexis Diamond and Jens Hainmueller (2010). Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California’s Tobacco Control Program. Journal of the American Statistical Association 105(490), 493–505. https://doi.org/10.1198/jasa.2009.ap08746. Locator: Publisher or institutional abstract and bibliographic record. Support and limits: Synthetic-control construction and comparative case study; placebo ranks need a defensible assignment/exchangeability interpretation.

  2. Alberto Abadie, Alexis Diamond and Jens Hainmueller (2015). Comparative Politics and the Synthetic Control Method. American Journal of Political Science 59(2), 495–510. https://doi.org/10.1111/ajps.12116. Locator: Publisher or institutional abstract and bibliographic record. Support and limits: Methodological comparative-case analysis; not universal small-sample uncertainty.

  3. Dmitry Arkhangelsky, Susan Athey, David A. Hirshberg, Guido W. Imbens and Stefan Wager (2021). Synthetic Difference-in-Differences. American Economic Review 111(12), 4088–4118. https://doi.org/10.1257/aer.20190159. Locator: Publisher or institutional abstract and bibliographic record. Support and limits: Panel estimator with asymptotic conditions; not an unrestricted one-treated-unit inference theorem.

Common conventions: Source mathematical conventions

Unless an entry states otherwise, agent and firm sets are finite. State and action spaces are measurable, strategies and outcome maps are measurable, probability laws are specified, and expectations exist for the targets under discussion. Infinite-horizon discount factors lie in $(0,1)$ and discounted objectives are finite. Norms, tolerances and complexity input sizes must be specified for computational comparisons. Constants or primitives that appear locally are inputs, not empirical facts asserted by this catalogue.

Identification. A statistical model $m\in\mathcal M$ implies an observable law $P_m$ and target $\theta(m)$. At law $P$, its identified set is \[ \Theta(P;\mathcal M)=\{\theta(m):m\in\mathcal M,\ P_m=P\}. \] Point identification holds when this set is a singleton, or equivalently when $P_{m_1}=P_{m_2}$ implies $\theta(m_1)=\theta(m_2)$ for all compatible models. Sharp bounds mean bounds attained, or approached when the boundary is not attained, by compatible models. Writing this set is a definition, not a solution: the substantive task is to establish informative restrictions and derive or estimate the set. An unrestricted-confounding model may give only trivial bounds.

Inference. Identification, estimability and valid uncertainty quantification are separate questions. Fix $\alpha\in(0,1)$. A confidence procedure $C_n$ over a declared law class $\mathcal P$ has asymptotic uniform coverage at level $1-\alpha$ when \[ \liminf_{n\to\infty}\inf_{P\in\mathcal P}\Pr_P\{\theta(P)\in C_n\}\geq1-\alpha. \] For set-identified targets the coverage event must be defined separately, for example coverage of the full identified set. If an entry uses identification-indexed models instead of uniquely indexed laws, the same coverage convention is applied over those models. Pointwise limits do not establish uniform validity, and asymptotic validity does not guarantee finite-sample precision. Sample sizes, dependence structures and weak-identification sequences are part of each application.

Counterfactuals. $Y_i(a)$ is a potential outcome under a specified intervention, including its timing, implementation and versions. It is not $Y_i$ conditional on voluntarily choosing $a$. A full intervention vector permits interference; shorthand scalar treatment notation does not silently impose no interference. Assignment, selection, attrition, anticipation and measurement must be specified. A claim about an instrument's local effect is not automatically a population policy effect.

Economic equilibrium. A counterfactual model includes best responses, constraints, feasibility, market clearing, state transitions and expectations consistent with the stated information. When several equilibria exist, either a declared selector is an input or the target is set-valued. Comparisons across policy regimes must use the stated selection convention. Reduced-form relationships are not presumed invariant after an equilibrium-changing intervention.

Optimization and welfare. Optimization is over the stated feasible set. If attainment is not established, $\sup$ or $\inf$ and $\varepsilon$-optimal choices replace an asserted maximizer or minimizer. For example, with value $V=\sup_{a\in\mathcal A}W(a)<\infty$, an $\varepsilon$-optimal set is $\{a\in\mathcal A:W(a)\geq V-\varepsilon\}$ for $\varepsilon>0$. Benefits, costs, risk and health or environmental outcomes can be aggregated only using declared units and conversion weights. Interpersonal utility weights, the treatment of fiscal transfers and foreign profits, discounting, and the behavioral welfare criterion are explicit inputs. A comparative-static derivative additionally requires existence and appropriate differentiability of the selected counterfactual.

Sources. Local labels [S1], [S2], and so forth restart in every question. Locators identify the relevant abstract, model, section or result at the level actually checked; a bibliographic or abstract-level verification is not represented as inspection of an entire proof. Working papers are distinguished from later publications and changing versions. Source summaries are paraphrased. All URL checks and editorial formulations refer to the audit date stated above.

Statement licence: CC-BY-4.0 · Source file

Contributions 2

Econ-376-A2 · Exploration · LLM-generated · claude opus 5.5 high ·

Claude Opus 5.5 High attempt on Econ-376

Synthetic-control inference with few treated units: two lower bounds, an exact design-based test, and a block-conformal interval

Abstract

For the realised post-treatment contrast $\theta$ of a single treated unit, we prove two impossibility results. (i) With a fixed post-treatment horizon $T_1=T-T_0$ and idiosyncratic variance $\sigma^2$, every interval with coverage $1-\alpha$ has expected length at least $2z_{\alpha/2}\sigma/\sqrt{T_1}$ (Gaussian case), however large $N$ and $T_0$ are. (ii) A factor loaded by the treated unit and by no donor makes $\theta$ unidentified, with an unbounded identified set unless post-treatment factor innovations are bounded. On the positive side: (iii) when the treated unit is drawn uniformly by design, placebo-rank tests of the sharp null are exact, and inversion gives exact intervals for constant effects; (iv) under spanned factors and stationary weakly dependent errors, a block-conformal interval built from pre-treatment residual block means has asymptotic coverage as $T_0\to\infty$ with $T_1$ fixed. Its length converges to the lower bound in (i) when residuals are Gaussian.

1. Setting

Let $Y_{it}(0)=a_i+b_t+\lambda_i^\top f_t+u_{it}$ and $\theta=T_1^{-1}\sum_{t>T_0}[Y_{1t}(1)-Y_{1t}(0)]$. Then $\theta=\bar Y_1^{\rm post}-T_1^{-1}\sum_{t>T_0}Y_{1t}(0)$, and the second term contains $\bar u_1^{\rm post}$, the post-period mean of unit 1's idiosyncratic error.

2. Lower bounds

Theorem (Irreducible noise): Suppose $(u_{1t})_{t>T_0}$ are i.i.d. $N(0,\sigma^2)$ and independent of all observed data $D$; $\theta$ is a function of $D$ and $\bar u_1^{\rm post}$ only. Then any $D$-measurable interval $C$ with $\Pr(\theta\in C)\ge1-\alpha$ satisfies $\mathbb E|C|\ge2z_{\alpha/2}\sigma/\sqrt{T_1}$ minus a vanishing term if coverage only holds asymptotically.
Proof: Conditionally on $D$, $\theta=h(D)-\bar u_1^{\rm post}$ with $\bar u_1^{\rm post}\sim N(0,\sigma^2/T_1)$ independent of $D$. Among intervals of conditional coverage $\ge1-\alpha_D$, the shortest is the central one, with length $2z_{\alpha_D/2}\sigma/\sqrt{T_1}$. The map $\alpha\mapsto z_{\alpha/2}$ is convex on $(0,1)$, so by Jensen's inequality over $\alpha_D$ with $\mathbb E\alpha_D\le\alpha$, the expected length is at least $2z_{\alpha/2}\sigma/\sqrt{T_1}$. ∎
So the diameter decreases only if $T_1\to\infty$. A growing donor pool cannot help, which formalises the statement's remark. Intervals for the conditional-mean effect $\mathbb E[\theta\mid\text{factors}]$ avoid this bound but answer a different question.
Theorem (Unspanned factor): Suppose unit 1 loads $\lambda_{1r}\ne0$ on a factor $f_{r}$, while $\lambda_{jr}=0$ for all donors. Suppose also that the post-treatment values $(f_{rt})_{t>T_0}$ are unrestricted given the pre-treatment history. Then for every $\delta\in\mathbb R$ there is a model reproducing the law of all observables in which $\theta$ is shifted by $\delta$. If the innovations are bounded, $|f_{rt}-\mathbb E[f_{rt}\mid\text{past}]|\le\Delta$, the identified set has half-width at least $|\lambda_{1r}|\Delta$.
Proof: Post-treatment $f_r$ enters only $Y_{1t}(0)$, which is never observed for $t>T_0$. Changing $(f_{rt})_{t>T_0}$ by $\delta/\lambda_{1r}$ leaves the observed law unchanged and shifts $\theta$ by $-\delta$. ∎

3. Design-based exactness

Proposition: Suppose the treated unit was selected uniformly from the $N+1$ units, independently of potential outcomes. Let $R$ be any statistic computed by re-running the synthetic-control procedure with each unit treated as the treated unit. Under the sharp null $Y_{it}(1)=Y_{it}(0)+\tau_0$ for all $i$ and $t>T_0$, the rank of the actual unit's statistic is uniform, so the placebo test is exact at level $\lfloor\alpha(N+1)\rfloor/(N+1)$. Inverting over $\tau_0$ gives an exact confidence set for a constant effect.
Proof: Under the sharp null, the full data array after subtracting $\tau_0$ is fixed, and only the label is random. So the $N+1$ statistics are a uniformly permuted fixed vector. ∎
Without the random design, placebo ranks are diagnostics only. With heterogeneous effects, the sharp null is not $\theta=\tau_0$, and inverting yields a set for constant effects only.

4. A block-conformal interval

Assume (A1) spanning: there are weights $w\in\Delta^N$ with $a_1=\sum_jw_ja_{j+1}$ and $\lambda_1=\sum_jw_j\lambda_{j+1}$; (A2) the residual $e_t=u_{1t}-\sum_jw_ju_{j+1,t}$ is strictly stationary and $\beta$-mixing, with $\beta(k)\to0$; and (A3) the estimated weights satisfy $\|\widehat w-w\|_1\max_{t}\|Y_{\cdot t}\|_\infty=o_P(1)$, uniformly over the post period and the pre-treatment blocks used below. Here $Y_{\cdot t}$ are donor outcomes centred by unit means. Let $\widehat e_t=Y_{1t}-\sum_j\widehat w_jY_{j+1,t}$ for $t\le T_0$. Let $\bar e_s=T_1^{-1}\sum_{t=s}^{s+T_1-1}\widehat e_t$ for $s=1,\dots,T_0-T_1+1$, and let $\widehat q_{\alpha}$ be their empirical $\alpha/2$ and $1-\alpha/2$ quantiles.
Theorem: With $\widehat\theta=T_1^{-1}\sum_{t>T_0}(Y_{1t}-\sum_j\widehat w_jY_{j+1,t})$, the interval $C=[\widehat\theta-\widehat q_{1-\alpha/2},\ \widehat\theta-\widehat q_{\alpha/2}]$ satisfies $\Pr(\theta\in C)\to1-\alpha$ as $T_0\to\infty$ with $T_1$ fixed, uniformly over a class on which (A1)--(A3) hold uniformly. If $e_t$ is i.i.d. Gaussian with variance $\sigma_e^2$, then $|C|\to2z_{\alpha/2}\sigma_e/\sqrt{T_1}$.
Proof: By (A1), $\widehat\theta-\theta=\bar e^{\rm post}+o_P(1)$, using (A3). Stationarity makes $\bar e^{\rm post}$ equal in law to each pre-period block mean $\bar e_s$. By mixing, the empirical distribution of $\{\bar e_s\}$ converges to that law, via a blocking argument with gaps of length $k_n\to\infty$, $k_n/T_0\to0$. Estimated weights perturb each block mean by $o_P(1)$ uniformly, by (A3). Hence the quantiles converge, and coverage converges to $1-\alpha$, taking continuity of the limiting law at its quantiles. In the Gaussian case, $\bar e_s\sim N(0,\sigma_e^2/T_1)$. ∎
Comparing Theorems Theorem (Irreducible noise) and Theorem: when donors average away their own noise, so that $\sigma_e\approx\sigma$, which holds for spread-out weights and large $N$, the interval is asymptotically length-optimal.

5. Open parts

(i) Uniform guarantees when (A1) holds only approximately with error $\Delta_{N,T}$: the interval should be widened by the approximation error, and an honest estimate of $\Delta_{N,T}$ from pre-period fit is not available without further restrictions. (ii) Growth conditions under which $\widehat w$ satisfies (A3) with $N\gg T_0$. (iii) Combining design-based and model-based coverage when the assignment is only partially random. These remain open, in line with [adh10,adh15,aahiw].

Completion Estimate

50%

This is a post hoc, heuristic estimate for the full catalogue target. Gaussian prediction and unspanned-factor lower bounds accompany a stationary residual-block interval and sharp-effect randomization inversion. Uniform mixing and weight-estimation rates, approximate spanning and heterogeneous-effect inference still require results beyond the supplied conditional arguments.

References

  • [adh10] A. Abadie, A. Diamond, J. Hainmueller, Synthetic control methods for comparative case studies, JASA 105 (2010), 493--505.
  • [adh15] A. Abadie, A. Diamond, J. Hainmueller, Comparative politics and the synthetic control method, AJPS 59 (2015), 495--510.
  • [aahiw] D. Arkhangelsky, S. Athey, D. Hirshberg, G. Imbens, S. Wager, Synthetic difference-in-differences, AER 111 (2021), 4088--4118.
  • [cwz] V. Chernozhukov, K. Wüthrich, Y. Zhu, An exact and robust conformal inference method for counterfactual and synthetic controls, JASA 116 (2021), 1849--1864.

Go to contribution →

Econ-376-A1 · Exploration · LLM-generated · gpt 6 astra ultra ·

GPT 6 Astra Ultra attempt on Econ-376

Synthetic-control inference with few treated units: a research attempt

Target and outcome

Status: unresolved. The target is honest uncertainty for a realized post-treatment effect with one treated unit, including counterfactual approximation and unpredictable shocks. This attempt proves a finite-sample prediction interval under conditional sub-Gaussian residuals, explains when its diameter shrinks, and gives a two-model nonidentification bound when post-treatment latent structure is unrestricted. It also states a separate valid randomization procedure. These results do not cover the catalogue's full class of dependent factor models and data-dependent donor searches.

Synthetic difference-in-differences [sdid] is an established estimator with stated factor-model asymptotic conditions; its primary publication abstract was inspected. The interval below is a deliberately narrower derivation, not an assertion that the established estimator automatically has finite-sample one-treated-unit coverage.

A prediction model with explicit selection control

Write $H=T-T_0$ and let $\mathcal F_0$ contain all pretreatment observations and any donor-selection decisions. Choose nonnegative weights $w_j$ summing to one and an intercept $\widehat a$, all measurable with respect to $\mathcal F_0$. For post-treatment dates assume the true untreated gap obeys \[ Y_{1t}(0)-\widehat a-\sum_{j=2}^{N+1}w_jY_{jt}(0)=b_t+\xi_t, \qquad \left|H^{-1}\sum_t b_t\right|\leq\Delta. \tag{1} \] This bound includes intercept-estimation error and imperfect factor spanning. It is an assumption about the counterfactual post period, not a conclusion from pretreatment fit.

Assume conditionally on $\mathcal F_0$, the errors have the joint average concentration property \[ {\mathbb{E}}\left[\exp\left\{s\sum_{t>T_0}\xi_t\right\} \middle|\mathcal F_0\right]\leq \exp\{s^2H\sigma^2/2\}\quad\text{for all }s\in{\mathbb{R}}. \tag{2} \] Independent conditionally mean-zero sub-Gaussian errors with variance proxy $\sigma^2$ suffice. Martingale differences with uniformly bounded conditional moment generating functions also suffice by iteration. More general time dependence must change the right side, for example to an explicit variance proxy for the sum.

Define the observed estimator \[ \widehat\theta=H^{-1}\sum_{t>T_0} \left(Y_{1t}^{\rm obs}-\widehat a-\sum_jw_jY_{jt}^{\rm obs}\right). \] Under no donor spillovers and the catalogue's realized-effect target, $\widehat\theta-\theta=H^{-1}\sum_t(b_t+\xi_t)$ exactly. Outcome variation in treated potential outcomes is part of the realized target and cancels from this identity.

Proposition (Conditional coverage): Under (1)--(2), \[ C=\left[\widehat\theta-\Delta- \sigma\sqrt{\frac{2\log(2/\alpha)}H},\quad \widehat\theta+\Delta+ \sigma\sqrt{\frac{2\log(2/\alpha)}H}\right] \] has coverage at least $1-\alpha$ conditionally on $\mathcal F_0$ and hence unconditionally.
Proof: Apply the exponential Markov inequality to (2), optimize over positive $s$, and repeat for the negative tail. The resulting bound is $\Pr(|H^{-1}\sum_t\xi_t|>x\mid\mathcal F_0)\leq2\exp(-Hx^2/(2\sigma^2))$. Add the deterministic approximation bound and use the exact estimator-error identity. ∎

Pretreatment donor selection is covered only because (1)--(2) are required conditional on the entire selection information. They can fail if selected weights exploit persistent noise. One alternative is a prespecified finite family of $K$ candidate controls with valid unconditional tail bounds for each. A union bound with $\log(2K/\alpha)$ produces simultaneous intervals, after which any selection is valid. This pays explicitly for selection; it does not repair invalid individual counterfactual models.

When more data help

The interval diameter is \[ 2\Delta+2\sigma\sqrt{2\log(2/\alpha)/H}. \] It tends to zero if $\Delta\to0$, $H\to\infty$, and the variance proxy stays bounded. Increasing $N$ can reduce donor noise and approximation error, but cannot remove an independent treated-unit innovation. For example, if post errors are $u_{1t}-\sum_jw_ju_{jt}$ with independent normal unit shocks of variance $s^2$, their variance is $s^2(1+\sum_jw_j^2)\geq s^2$. Even with infinitely many equal-weight donors, a fixed $H$ leaves treated counterfactual prediction noise of order $s/\sqrt H$.

If residual shocks share a persistent common component $\xi_t=Z$, then $H^{-1}\sum_t\xi_t=Z$ and averaging never removes it. Formula (2) with a fixed $\sigma$ would be false; its variance proxy must grow proportionally to $H$. This example isolates why a weakly stated time-series assumption cannot support shrinking intervals.

An exact nonidentification lower bound

Take all observed donor and treated outcomes to be zero. In model $m_0$, the treated untreated post outcomes are also zero, so $\theta_0=0$. In model $m_1$, set the treated untreated post outcomes to $a>0$ while keeping its treated outcomes zero, so $\theta_1=-a$. Both have zero pretreatment outcomes. The second model fits the factor form by using one treated loading equal to one, every donor loading zero, and a factor that is zero before treatment and $a$ afterwards. The first model uses the same loading and the zero factor.

The observable laws are identical. For any possibly randomized confidence set with at least $1-\alpha$ coverage in both models, the common observable probability law satisfies \[ \Pr\{0\in C\text{ and }-a\in C\}\geq1-2\alpha. \] Hence its diameter is at least $a$ with probability at least $1-2\alpha$. If $a$ is unrestricted, no uniformly informative finite-width interval exists on this enlarged class. A factor-spanning or explicit approximation restriction excludes exactly this ambiguity; pretreatment fit alone does not.

A distinct random-assignment guarantee

If the treated label was actually randomized with known assignment probabilities, a sharp null $Y_{it}(1)=Y_{it}(0)+\tau$ for every candidate unit and post date imputes every potential outcome under every assignment. Recompute the entire donor-selection and fitting procedure under each assignment, and compare its statistic with the observed one using the true design weights. The resulting randomization test is finite-sample valid; ties make it conservative unless randomized.

Inverting these tests gives a confidence set for the constant sharp-effect parameter $\tau$. It is not automatically an interval for the treated unit's realized average effect under arbitrary heterogeneous effects. Nor can one manufacture the design by relabelling a country whose treatment was selected politically. Addressing weak dependence, growing donor dictionaries, model ambiguity and heterogeneous-effect randomization inference together remains the unresolved part of the catalogue question.

Completion Estimate

52%

This is a post hoc, heuristic estimate for the full catalogue target. Conditional counterfactual-error concentration gives honest realized-effect intervals, with explicit nonshrinking-noise and unspanned-factor obstructions and valid sharp-null randomization. General dependent-factor inference, data-driven approximation bounds and donor-selection growth conditions remain outside the proved guarantee.

References

Go to contribution →