RESEARCH
Job Market Paper
Sufficient Instruments Filter
This paper introduces a novel procedure to filter out sufficient information from many instruments for the estimation of parameters in regression models with endogenous regressors. This method allows correlated and even invalid instruments. This technique draws its merit from the ability to incorporate supervision, the flexibility to accommodate non-linearity, and the capability for sufficient dimension reduction.
[Paper]
Abstract
This paper introduces a novel five-layered deep learning-based tractable procedure to filter out sufficient information from many instruments for estimating parameters in regression models with endogenous regressors. The method draws its merit from three key properties: the ability to incorporate supervision, the flexibility to accommodate non-linearity, and the capability for sufficient dimension reduction. This method is consistent and asymptotically normal when many instruments are correlated. Simulation exercises show that this method consistently achieves lower bias and root mean squared error compared to competing benchmarks, across many specifications. Two real-world applications in industrial organizations(IO) and finance are considered, yielding meaningful insights into causal relationships. The method remains robust when the number of instruments exceeds the sample size, and performs well with weak and even invalid observed instruments, as long as there exists at least one linear combination of common factors among the observed instruments that serves as a valid instrument.
Working Papers
Kernel Three Pass Regression Filter
(with Daanish Padha)
[Accepted for Presentation in The 2024 California Econometrics Conference]
[Accepted for Presentation in European Winter Meeting of the Econometric Society (EWMES) 2024]
[Accepted for Presentation in 19th Annual Conference on Economic Growth and Development at Indian Statistical Institute]
[Accepted for Presentation in The 34th Annual Midwest Econometrics Group Conference, USA]
(with Daanish Padha)
[Accepted for Presentation in The 2024 California Econometrics Conference]
[Accepted for Presentation in European Winter Meeting of the Econometric Society (EWMES) 2024]
[Accepted for Presentation in 19th Annual Conference on Economic Growth and Development at Indian Statistical Institute]
[Accepted for Presentation in The 34th Annual Midwest Econometrics Group Conference, USA]
We propose a novel supervised and non-linear method of forecasting a single time series using a high-dimensional set of predictors. The method is computationally efficient and demonstrates strong empirical performance, particularly over longer forecast horizons.
[Paper] [Poster]
Abstract
We forecast a single time series using a high-dimensional set of predictors. When predictors share common underlying dynamics, a latent factor model estimated by the Principal Component method effectively characterizes their comovements. These latent factors succinctly summarize the data and aid in prediction, mitigating the curse of dimensionality. However, two significant drawbacks arise: (1) not all factors may be relevant, and utilizing all of them in constructing forecasts leads to inefficiency, and (2) typical models assume a linear dependence of the target on the set of predictors, which limits accuracy. We address these issues through a novel method: Kernel Three-Pass Regression Filter. This method extends a supervised forecasting technique, the Three-Pass Regression Filter, to exclude irrelevant information and operate within an enhanced framework capable of handling nonlinear dependencies. Our method is computationally efficient and demonstrates strong empirical performance, particularly over longer forecast horizons.
The Agricultural Productivity Gap: Informality Matters
(with Bharat Ramaswami)
[Under Review in the Journal of Development Economics.]
[Media Coverage by Ideas for India.]
(with Bharat Ramaswami)
[Under Review in the Journal of Development Economics.]
[Media Coverage by Ideas for India.]
1 / 3
Productivity Gap --- Informality
2 / 3
Informed View of Productivity Gap
3 / 3
Policy Implication
We find that the primary dualism in development is between the formal non-farm sector and the informal sector including agriculture. Non-parametric econometric techniques are used for analysis.
[Paper]
Abstract
The literature has debated whether the productivity gap between agriculture and non-agriculture reflects mobility barriers or selection. Non-agriculture is not a homogeneous category. In developing countries, most of the non-agricultural employment is informal. Could it be that the productivity gap is driven by formal sector firms that are numerically small but economically substantial? This paper compares the productivity of agriculture to the informal and formal non-farm sectors in India. The comparison controls for sectoral differences in hours worked, human capital, and labor share of value added. The paper finds substantial productivity gaps with the formal sector but small and negligible gaps with the informal non-farm sector. Between 40-50% of non-farm workers are in sectors not more productive than agriculture. These findings suggest that the primary dualism in development is between the formal non-farm sector and the informal sector including agriculture.
Work In Progress
Supervised Deep Factor Models
(with Daanish Padha)
(with Daanish Padha)
We employ supervision in deep learning methods to learn the underlying low-rank latent structure of the data. Then use this learning to forecast the out-of-sample time series. This method promises strong empirical performance.
[Draft Coming Soon]
Abstract
We use a neural network to forecast a single time series. Inspired by the "Targeted Predictors" approach from Bai (Journal of Econometrics, 2008), we first select a set of predictors by performing polynomial regression for each predictor individually. Unlike traditional factor models, which limit the search to an underlying planar structure, our approach explores a non-linear, low-dimensional representation of the predictors that best explain the target variable.
Information Theoretic Maximum Entropy Density Estimator
(with Amos Golan, Tae-Hwy Lee, Millie Mao, and Aman Ullah)
(with Amos Golan, Tae-Hwy Lee, Millie Mao, and Aman Ullah)
Developing a new distribution learning method for faster non-parametric estimations. Unlike local variation-based kernel-based non-parametric density estimator, this method is global, which makes it faster.
[Draft Coming Soon]
Abstract
Existing non-parametric kernel density estimators struggle to accurately estimate density in tails, especially in the fat-tailed distributions. In this work, we aim to develop a multivariate maximum entropy (ME) density estimator by matching sample moments with population moments. The method shows promising results beating kernel-based density estimators in tails and overall in one-dimensional density estimation problems.
Presentations in Research Conferences/Seminars
| Dec 2024: The European Winter Meeting of the Econometric Society (EWMES 2024) | Palma, Spain |
| Nov 2024: 34th Annual Midwest Econometrics Group Conference at Uni. of Kentucky | Lexington, KY, USA |
| Oct 2024: Fall 2024 Econometrics Seminar at UC Riverside | Riverside, CA, USA |
| Sep 2024: The 2024 California Econometric Conference at UC Davis | Davis, CA, USA |
| Oct 2023: Fall 2023 Econometrics Seminar at UC Riverside | Riverside, CA, USA |
| May 2023: Spring 2023 Brown Bag Seminar at UC Riverside | Riverside, CA, USA |
| Feb 2023: Winter 2023 Brown Bag Seminar at UC Riverside | Riverside, CA, USA |
| Dec 2022: Annual Conference by The Econometric Society & Delhi School of Economics | Delhi, India |
| Dec 2019: Annual Conference by the Indian Statistical Institute | Delhi, India |