# High-dimensional and sparse IV methods

> One thread of 6 from the "instrumental variable" map, covering the 11 papers in it. The other threads are not represented here.
>
> Every quotation was copied word for word from the paper's own text, and
> checked against that text. Quotes marked *inferred* failed that check and
> must be re-checked before use. Quotes marked *not re-checked* have not been
> matched against the paper's text as it now stands, so they carry no current
> verification either. Lines labelled *the tool's reading* are
> model judgment, not quotation, and carry no verification.
>
> **This is a scaffold, not prose.** The citations, quotes and structure are
> real; the argument is yours to write.

> Connects to: Classical IV foundations and refinement, IV with heterogeneous treatment effects and causal mechanisms, Specialized IV applications and domain extensions.

This line began with LASSO-based first-stage selection in high-dimensional IV models and grew to encompass Post-LASSO, sqrt-LASSO, and sparsity-aware inference. Work here targets settings where the number of potential instruments or control variables is large, using regularization to maintain consistency and asymptotic normality while controlling model dimension.


**Sparse Models and Methods for Optimal Instruments with an Application to Eminent Domain** (2010) \cite{belloni2010sparse}

What it did: Establish root-n consistency for Lasso-based IV estimation  *(the tool's reading)*

This paper introduces the first theoretical guarantees that a Lasso/Post-Lasso estimator can achieve root-n consistency and asymptotic normality for a low-dimensional structural parameter when the instrument set is high-dimensional, without requiring sparsity constraints on the structural parameter itself.  *(the tool's reading)*

> “The model is y_i = d_i'α_0 + ϵ_i where α_0 denotes the true value of a vector-valued parameter α. y_i is the response variable, and d_i is a finite k_d-vector of variables whose first k_e elements contain endogenous variables. The disturbance ϵ_i obeys for all i (and n): E[ϵ_i|x_i] = 0, where x_i is a k_x-vector of instrumental variables.”
>
> ✓ verified: found word for word in the paper's own text


**LASSO Methods for Gaussian Instrumental Variables Models** (2010) \cite{belloni2010lasso}

What it did: Apply sparse methods to construct optimal instruments  *(the tool's reading)*

Building on Lasso-based consistency, this paper demonstrates how sparse prediction methods can form the optimal instruments in the first stage of linear IV models, extending the approach to cases with many instruments where traditional methods fail.  *(the tool's reading)*

> “Identification of the causal effects of interest in this setting may be achieved through the use of observed instrumental variables that are relevant in determining the treatment status but are otherwise unrelated to the outcome of interest.”
>
> ✓ verified: found word for word in the paper's own text


**A new selection method for high-dimensionial instrumental setting: application to the Growth Rate convergence hypothesis** (2011) \cite{mougeot2011new}

What it did: Automate instrument selection from large candidate sets  *(the tool's reading)*

This paper shifts the focus from constructing optimal instruments to selecting relevant instruments objectively from a huge set of candidates, embedding the IV problem in a very high-dimensional setting to avoid ad-hoc variable specification.  *(the tool's reading)*

> “The insertion of instrumental variables Z in the model may lead to consistent estimation of the coefficients α.”
>
> ✓ verified: found word for word in the paper's own text


**Inference for High-Dimensional Sparse Econometric Models** (2011) \cite{belloni2011inference}

What it did: Develop inference theory for high-dimensional sparse IV models  *(the tool's reading)*

This paper extends the high-dimensional sparse regression framework to establish novel inference results specifically for IV models with many instruments, providing valid confidence intervals and hypothesis tests in this setting.  *(the tool's reading)*

> “We then develop HDS models and methods in instrumental variables models with many instruments in Section […] and a partially linear model with many series terms in Section […], with the main emphasis given to inference.”
>
> ✓ verified: found word for word in the paper's own text


**Shrinkage priors for linear instrumental variable models with many instruments** (2014) \cite{hahn2014shrinkage}

What it did: Introduce shrinkage priors for high-dimensional first stages  *(the tool's reading)*

This paper develops a predictor-dependent shrinkage prior based on the Frisch decomposition to regularize the first-stage regression in the many-instruments setting, offering a Bayesian alternative to Lasso-based approaches.  *(the tool's reading)*

> “The variable x_i is referred to as the treatment variable, y_i is the response variable and z_i is a vector of instruments.”
>
> ✓ verified: found word for word in the paper's own text


**Bias Reduction in Instrumental Variable Estimation through First-Stage\n Shrinkage** (2017) \cite{spiess2017bias}

What it did: Combine James-Stein shrinkage with control-function approaches  *(the tool's reading)*

This paper applies James-Stein type shrinkage to the first-stage high-dimensional normal-means problem and follows with a control-function method in the second stage, providing an alternative shrinkage strategy for IV estimation.  *(the tool's reading)*

> “from n iid observations (Y_i,X_i,Z_i,W_i), where X_i ∈ℝ is the regressor of interest (assumed univariate), W_i ∈ℝ^k control variables, Z_i ∈ℝ^ℓ instrumental variables, and (U_i,V_i)' ∈ℝ^2 is homoscedastic (wrt Z_i), Normal noise.”
>
> ✓ verified: found word for word in the paper's own text


**LASSO-Driven Inference in Time and Space** (2018) \cite{chernozhukov2018lasso}

What it did: Embed IV de-biasing into LASSO inference for dependent data  *(the tool's reading)*

This paper uses an IV-based de-biasing technique as a technical device within a LASSO inference framework that handles temporal and cross-sectional dependence, extending sparse IV methods beyond independent cross-sectional settings.  *(the tool's reading)*

> “Run LS IV regression of Y_j,t - X_j(-k),t^⊤β^[1]_j(-k) on X_jk,t using v_jk,t as an instrument variable, attaining the final estimator β^[2]_jk.”
>
> ✓ verified: found word for word in the paper's own text


**High-Dimensional Mixed-Frequency IV Regression** (2020) \cite{babii2020high}

What it did: Relax order condition using low-frequency instrumental variables  *(the tool's reading)*

This paper shows that identification and estimation of high-dimensional slope parameters is possible with a single low-frequency instrumental variable, relaxing the traditional requirement that the number of instruments equals or exceeds the number of endogenous regressors.  *(the tool's reading)*

> “We show that the high-dimensional slope parameter of a high-frequency covariate can be identified and accurately estimated leveraging on a low-frequency instrumental variable.”
>
> ✓ verified: found word for word in the paper's own text


**Inference on the New Keynesian Phillips Curve with Very Many Instrumental Variables** (2021) \cite{dovi2021inference}

What it did: Test weak identification with many time-series instruments  *(the tool's reading)*

This paper develops a Sup Score test for IV-based inference that remains valid under dependent data, arbitrarily weak identification, and a number of instruments that grows with sample size, extending valid inference to time-series settings.  *(the tool's reading)*

> “In virtually all applications, the relation is assumed to contain an additive error term that is shown (e.g., by the assumption of Rational Expectations (RE)) or primitively assumed to be uncorrelated with predetermined variables excluded from the specified relation. This makes any predetermined variable a valid IV.”
>
> ✓ verified: found word for word in the paper's own text


**A Distance Covariance-based Estimator** (2021) \cite{tsyawo2021distance}

What it did: Relax IV relevance condition using distance covariance  *(the tool's reading)*

This paper introduces a new estimator that relaxes the conventional linear-correlation-based IV relevance condition, allowing endogenous covariates to be weakly or nonlinearly correlated with instruments while maintaining identification.  *(the tool's reading)*

> “[…] is the condition of non-independence between non-trivial linear combinations of X and Z; it is the MDep analogue of the relevance condition in the IV setting, e.g., […], and an MDep analogue of the linear completeness condition in ICM estimators, e.g., […]. In the IV setting, the relevance condition requires that no non-zero linear combination of X be uncorrelated with Z.”
>
> ✓ verified: found word for word in the paper's own text


**On the instrumental variable estimation with many weak and invalid instruments** (2022) \cite{lin2022instrumental}

What it did: Handle weak and invalid instruments simultaneously  *(the tool's reading)*

This paper proposes a sparse-rule-based estimator that identifies treatment effects when instruments are both weak and potentially invalid, using sparsity to distinguish between relevant and irrelevant invalid instruments.  *(the tool's reading)*

> “The instrumental variable (IV) method is widely used when the treatment variable of interest is endogenous. As shown in Figure […], the ideal IV needs to be correlated with the endogenous treatment variable (C1), it should not have a direct effect on the outcome (C2) and should not be related to unobserved confounders that affect both outcome and treatment (C3).”
>
> ✓ verified: found word for word in the paper's own text


_[Your synthesis: what it enabled, what it left unsolved.]_


## Where this thread connects

Each crossing is where one line of work fed another. These are the tool's reading of
the corpus, not quotations.

- **Classical IV foundations and refinement** → **High-dimensional and sparse IV methods** (2010): Many instruments, LASSO first-stage
- **High-dimensional and sparse IV methods** → **IV with heterogeneous treatment effects and causal mechanisms** (2019): Machine learning heterogeneous effects
- **High-dimensional and sparse IV methods** → **Specialized IV applications and domain extensions** (2020): Mixed-frequency IV, panels