- 1.1.2. Probabilistic Bounds. We start by lower bounding the probability of B. By Assumption A.4(a), we have that kX2 j,tkÏp/2 ⤠C for some constant depending only on Ïp. For the polinomial case it follows from Cauchy-Schawartz inequality. For the exponential case it follows from Lemma C.18 in Fan, Masini and Medeiros (2021a).
Paper not yet in RePEc: Add citation now
Abadie, Alberto (2021) âUsing synthetic controls: Feasibility, data requirements, and methodological aspects,â Journal of Economic Literature, 59 (2), 391â425.
Abadie, Alberto and Jeremy LâHour (2021) âA Penalized Synthetic Control Estimator for Disaggregated Data,â Journal of the American Statistical Association, 116 (536), 1817â1834.
- Abadie, Alberto and Matias D Cattaneo (2018) âEconometric methods for program evaluation,â Annual Review of Economics, 10, 465â503.
Paper not yet in RePEc: Add citation now
Acemoglu, Daron, Simon Johnson, Amir Kermani, James Kwak, and Todd Mitton (2016) âThe value of connections in turbulent times: Evidence from the United States,â Journal of Financial Economics, 121 (2), 368â391.
Angrist, Joshua, Victor Chernozhukov, and IvaÌn FernaÌndez-Val (2006) âQuantile Regression under Misspecification, with an Application to the U.S. Wage Structure, â Econometrica, 74 (2), 539â563.
- Belloni, Alexandre and Victor Chernozhukov (2011) âl1-penalized quantile regression in high-dimensional sparse models,â The Annals of Statistics, 39 (1), 82â130.
Paper not yet in RePEc: Add citation now
- Belloni, Alexandre, Victor Chernozhukov, and Christian Hansen (2014) âHighDimensional Methods and Inference on Structural and Treatment Effects,â Journal of Economic Perspectives, 28 (2), 29â50.
Paper not yet in RePEc: Add citation now
Carvalho, Carlos V., Ricardo Masini, and Marcelo C. Medeiros (2018) âArCo: An artificial counterfactual approach for high-dimensional panel time-series data,â Journal of Econometrics, 207 (2), 352â380.
Chen, Yi-Ting (2020) âA distributional synthetic control method for policy evaluation, â Journal of Applied Econometrics, 35 (5), 505â525.
Chernozhukov, Victor and Christian Hansen (2005) âAn IV model of quantile treatment effects,â Econometrica, 73, 245â261.
Chernozhukov, Victor, IvaÌn FernaÌndez-Val, and Alfred Galichon (2010) âQuantile and probability curves without crossing,â Econometrica, 78 (3), 1093â1125.
- Conditional on X1, . . . , XT0 and U1, . . . UT0 , the process {T â1/2 PT0 t=1 W0 t,j(Ï) : Ï â T } is Sub-Gaussian with respect to the semi-metric kkPn,2. Let Pε be the law with respect to the Rademacher random variables and kke2,Pε denote the Ï-Orlicz norm with Ï(x) = exp x2 + 1 respect to Pε. Then for any Ï0 â T , by Corollary 2.2.8 in Van Der Vaart and Wellner (1996), kT â1/2 0 S0 j ke2,Pε ⤠kT â1/2 T0 X t=1 εtWt,j(Ï0)ke2,Pε + C ⤠2 Tâ1 T0 X t=1 W2 t,j(Ï0) !1/2 + C ⤠2 + C =: C1 Then, for x > 0, P(S0 j > x â T0) ⤠2 exp[â(x/C1)2 ] by Lemma C.17 in Fan, Masini and Medeiros (2021a). Set x = 4ÏC1 â T0 log n for Ï â¥ 1 to conclude that P(Sj > x) ⤠4EεPε(S0 j > x/4) ⤠8nâÏ2 , and therefore P(A ) ⥠1 â 8nâÏ2+1 for λ0 := 4C1Ï q log n T0 and Ï â¥ 1.
Paper not yet in RePEc: Add citation now
- Define the function class F := {(x, u) 7â x(1{u ⤠Ï} â Ï)/vj : Ï â T }. Since F is has VC dimension at most 6 with envelope function F(x) := |x|/vj, Lemma 2.6.7 in Van Der Vaart and Wellner (1996) ensures that N(kFkQ,2, F, L2(Q)) ⤠Câ10 for some constant C, for all â (0, 1), and any probability measure Q. Then, kFkPn,2 Z 1 q log N(kFkQ,2, F, L2(Q))d ⤠1 à C.
Paper not yet in RePEc: Add citation now
- DISTRIBUTIONAL COUNTERFACTUAL ANALYSIS 3 We now bound the probability of A . Let Aj := {supÏâT |Tâ1 PT0 t=1 Wt,j(Ï)| ⤠λ0} for 1 ⤠j ⤠n and λ0 > 0. Then A = Tn j=1 Aj. By the union bound P(A c ) = P( Sn j=1 A c j ) ⤠n maxj P(A c j ). Let W0 t,j(Ï) = εtWt,j for 1 ⤠t ⤠T0, 1 ⤠j ⤠n and Ï â T , where ε1, . . . εT0 is an iid sequence of Rademacher random variables independent of X1, . . . , XT0 and U1, . . . UT0 . Let Sj := supÏâT | PT0 t=1 Wt,j(Ï)|. By a symmetrization argument, Lemma 2.3.7 in Van Der Vaart and Wellner (1996) , we have P(Sj > x) ⤠4P(S0 j > x/4); x ⥠â 8C where S0 j is defined as Sj but with Wt,j replaced by the W0 t,j.
Paper not yet in RePEc: Add citation now
- Fan, Jianqing, Ricardo Masini, and Marcelo C. Medeiros (2021a) âBridging factor and sparse models.â (2021b) âDo we exploit all information for counterfactual analysis? benefits of factor models and idiosyncratic correction,â Journal of the American Statistical Association, 1 (forthcoming), 1â53.
Paper not yet in RePEc: Add citation now
- Feng, Yingjie (2021) âCausal Inference in Possibly Nonlinear Factor Models.â He, Xuming, Lan Wang, and Hyokyoung Grace Hong (2013) âQuantile-adaptive model-free variable screening for high-dimensional heterogeneous data,â The Annals of Statistics, 41 (1), 342 â 369, 10.1214/13-AOS1087.
Paper not yet in RePEc: Add citation now
Koenker, Roger, Samantha Leorato, and Franco Peracchi (2013) âDistributional vs. quantile regression,â CEIS Working Paper No 300, 1 (1).
- Talagrand, Michel (1989) âIsoperimetry and integrability of the sum of independent Banach-space valued random variables,â The Annals of Probability, 1546â1570.
Paper not yet in RePEc: Add citation now
- Then k PT0 t=1 X2 j,t â EX2 j,tkÏp/2 ⤠C â T0. The polinominal case follows from the Marcinkiewicz-Zygmund inequality while the exponential case can be derived from Theorem 3 and 4 in Talagrand (1989). By Markovâs inequality, for x > 0: P(|vj â Evj| > x) ⤠Ïp/2(x â T0/C) . Hence, using A.4(b) and the union bound we conclude that P(B) ⥠P(kv â Evkâ ⤠C/2) = 1 â P(kv â Evkâ > C/2) ⥠1 â n max 1â¤jâ¤n P(|vj â Evj| > C/2) ⥠1 â n Ïp/2(C1 â T0) .
Paper not yet in RePEc: Add citation now
- Van Der Vaart, Aad and Jon Wellner (1996) Weak convergence and empirical processes: with applications to statistics: Springer Science & Business Media. DISTRIBUTIONAL COUNTERFACTUAL ANALYSIS 1 Supplemental Material: Distributional Counterfactual Analysis in High-Dimensional Setup This supplemental material is organized in two sections. The first one contains the proofs of Theorem 1 and 2 stated in Section 3.2 of the main text. The second section collects additional figures from the empirical exercises described in Section 5 of the main text.
Paper not yet in RePEc: Add citation now