Arka P. Bandyopadhyay
Portrait of Arka P. Bandyopadhyay
A simulated time series in gold and its time reversal dashed in amber
SeriesTime-reversed

When a series and its time reversal are statistically distinguishable, the arrow of time reveals what drives what — the idea behind my work on directional irreversibility.

Arka P. Bandyopadhyay

Finance, real estate and AI · Adjunct Professor at Columbia, NYU and Yeshiva

I lead research that moves from rigorous finance to decisions that matter — in mortgage and real estate markets, and in how institutions put AI to work.

Former Head of AI and Quantitative ResearchFranklin Templeton, private equity and CRE
Published in JFQA, JEDC and QJFHousehold finance, mortgages, AI
Over a decade in industryRocktop, Nuveen/TIAA, UBS, Santander, Deloitte

Profile

My career has run on two tracks at once: research that meets the standards of top finance journals, and leadership of teams that turn that research into systems institutions rely on.

As an academic, I study household and consumer finance, mortgage markets, real estate and AI in finance, with publications in the Journal of Financial and Quantitative Analysis, the Journal of Economic Dynamics and Control and the Quarterly Journal of Finance. I teach finance, real estate and quantitative methods at Columbia, NYU and Yeshiva University, and previously held faculty appointments at the University of Miami and CU Boulder.

In industry, I led AI and quantitative research at Franklin Templeton in private equity and commercial real estate, one of the largest U.S. real estate investment managers, building agentic AI, LLM and retrieval systems and owning AI in production. Before that I spent a decade in mortgage credit, cross-asset research, loss forecasting and structured credit at Rocktop Partners, Nuveen/TIAA, UBS, Santander and Deloitte.

  • Household and consumer finance
  • Real estate
  • Mortgage markets
  • AI in finance
  • Banking and FinTech
  • Asset pricing from narrative
  • Applied econometrics

Research

Selected work, with links to each paper on SSRN. Full list and earlier versions on the SSRN author page.

Job market paper

Refereed publications

With Dongshin Kim and Patrick S. Smith · Journal of Financial and Quantitative Analysis, 2026, 61(3), 1148–1177
Abstract

Ginnie Mae issuers may buy seriously delinquent loans out of MBS pools at par. Conditional on default, a one-percentage-point higher interest rate spread raises the probability of an early buyout by 7–9 percentage points: issuers select the loans most valuable when they reperform, at the expense of MBS investors, and buyout activity depends on issuers' access to capital.

+7–9 pp per 1 pp of spread interest rate spread on the loan P(early buyout | default)
Issuers buy out the high-spread loans that are most valuable if they reperform — at MBS investors' expense.
With Lilia Maliar · Journal of Economic Dynamics and Control, 2026, 182, 105229
Abstract

Model-free reinforcement learning is used to derive how a mortgage servicer should act toward a borrower, replacing heuristic industry rules. Using post-securitization soft information and a new measure of borrower responsiveness, the servicer learns the borrower's type dynamically, anticipating strategic behavior and raising cooperation.

Servicer RL policy Borrower unknown type action response responsiveness → belief learn the type, then act
The servicer learns each borrower's type from how they respond, replacing fixed industry rules with a learned policy.

Papers under review

Solo-authored Under review
Summary

Develops the Directional Irreversibility Index (DII), which tests whether the joint distribution of an identified shock and a later financial outcome favors the shock-to-outcome representation over the reverse — something mean and volatility responses cannot tell us. Shows what bounds or erases the signal, why a naive calibration over-rejects, and how a joint bootstrap restores size while removing common factors raises power.

X Y shocks travel this way not this way t → ← −t
Run the series backwards and it no longer looks the same: that asymmetry says which way shocks travel.
Solo-authored Under review
Summary

Directional dependence diagnostics infer that one series drives another from an asymmetry between forward and backward residual dependence — but an unobserved state tracked nonlinearly by both series produces the same asymmetry. The paper bounds directional indices under such confounding and computes a robustness value: the minimum loading a confounding explanation would need. Applied to 2,000 central-bank communication events, equity-index asymmetries turn out not to establish the direction of monetary transmission.

no longer distinguishable robustness value finding overturned strength of hidden confounder directional evidence
How strong would a hidden common driver have to be to explain the asymmetry away? The robustness value puts a number on it.
Solo-authored Under review
Summary

A liquidity constraint can shut down the transaction designed to relieve it. Ginnie Mae issuers must advance payments on non-paying loans; the early buyout is the only contractual escape, yet exercising it consumes the balance sheet the obligation drains. Across 3.1 million buyout decisions by 385 issuers, nonbank buyout rates fell from 45% to 10% when COVID-19 struck while depository rates rose from 80% to 92%, steepest for the highest-coupon loans — balance sheets, not option values, decided which borrowers were bought out.

before COVID-19 0% 100% 92% 80% 10% 45% depositories nonbanks buyout rate on eligible delinquent loans
Same option, opposite responses: banks bought out more when the crisis hit; balance-sheet-constrained nonbanks nearly stopped.
Solo-authored Under review
Summary

The CARES Act paired forbearance with a foreclosure moratorium for federally backed loans, making the two policies hard to separate. Using 12,691 loans from one servicer, the paper exploits private-label mortgages that received neither protection but could be granted discretionary forbearance. Among them, an April 2020 forbearance conversation predicts a 9 to 15 percentage point increase in subsequent modification, with reduced adverse resolution concentrated among borrowers already close to foreclosure.

+9–15 pp no forbearance conversation forbearance conversation, Apr 2020 later modification rate private-label loans outside the CARES Act perimeter
Outside the statutory perimeter, a forbearance conversation alone shifted loans away from foreclosure and toward modification.
Solo-authored Under review
Summary

Housing demand from migration depends on incoming households' purchasing power, which is usually unknown in advance. Using the Social Connectedness Index, the paper measures each county's exposure to house-price gains in the distant counties it is socially tied to. That exposure predicts the income of migrants arriving years later, outperforms short-term appreciation signals, and it is the composition of arrivals — not their number — that predicts destination price appreciation.

equity equity accumulated in a connected market — social ties (SCI) → arriving purchasing power
Housing wealth built in one market reappears as the purchasing power of households arriving somewhere else, along social ties.

Working papers

Leadership and industry

Building research functions and putting models and AI into production.

  • 2024 – 2026
    Franklin Templeton, private equity and commercial real estate
    Head of AI and Quantitative Research

    Led AI and quantitative research at one of the largest U.S. real estate investment managers. Built agentic AI, LLM/RAG and knowledge-graph document systems, and owned AI in production.

  • 2019 – 2021
    Rocktop Partners
    Principal Data Scientist

    Mortgage credit and Ginnie Mae early-buyout models.

  • 2018 – 2019
    Nuveen / TIAA
    Investment Quantitative Research

    Cross-asset quantitative research.

  • 2013 – 2018
    UBS, Santander Bank, Deloitte & Touche
    Quantitative risk and structured credit

    Loss forecasting, structured credit and derivatives.

Languages
Python, R, SAS, Stata, MATLAB, VBA
AI and ML
PyTorch, LangChain, LangGraph; LLMs, RAG, agentic AI, knowledge graphs
Methods
Deep learning, NLP on financial text, reinforcement learning, causal inference, panel econometrics
Data
MS SQL, MongoDB, KDB+, vector databases; Flask, Docker, Kubernetes, Airflow

Teaching

Undergraduate, MBA, MSF and MSRE programs.

  • 2023 – present
    Columbia University, New York University, Yeshiva University
    Adjunct Professor, Finance, Economics and Real Estate
  • 2022 – 2023
    University of Miami, Herbert Business School
    Professor of Professional Practice, Finance and Real Estate
  • 2021 – 2022
    University of Colorado Boulder, Leeds School of Business
    Scholar-in-Residence, Finance and Real Estate

Finance and quantitative methods

  • Financial EconomicsColumbia
  • Financial Risk and AnalyticsNYU Tandon, Yeshiva
  • Options and FuturesCU Boulder, Yeshiva
  • Principles of FinanceCU Boulder, Miami, Yeshiva
  • Decision ModelingCU Boulder, Yeshiva
  • Quantitative ModelingYeshiva

Real estate

  • Real Estate FinanceCU Boulder, Yeshiva
  • Real Estate RiskCU Boulder, Miami, NYU
  • Data Analytics for Real EstateNYU Schack
  • Statistics Applications in Finance and Real EstateColumbia SPS, NYU Schack

Education and honors

  • 2018 – 2021
    Ph.D., Finance, Real Estate and Artificial Intelligence
    Baruch College, Zicklin School of Business, CUNY
  • 2011 – 2013
    ABD, Applied Mathematics; M.S., Mathematical Finance
    Courant Institute of Mathematical Sciences, New York University
  • 2010 – 2011
    M.S., Computer Science
    Louisiana State University
  • 2005 – 2008
    B.Math. (Honors), Mathematics
    Indian Statistical Institute, Bangalore
  • Brattle Ph.D. Group Award, Western Finance Association, 2022 — one of 11 Ph.D.s selected globally
  • AREUEA Dissertation Award, 2020
  • American Finance Association Doctoral Student Grant, 2020
  • PSC-CUNY Research Grant, with Lilia Maliar

Contact

Email
ab3985@columbia.edu, apb321@nyu.edu
Location
New York area. U.S. Permanent Resident.
Profiles
SSRN, Google Scholar