About

Neil Shephard is the Frank B. Baird Jr. Professor of Science at Harvard University, holding joint appointments as Professor of Economics and Professor of Statistics. His broad research interests lie in econometrics, finance, and statistics, with a particular focus on financial econometrics. He has made significant advances in developing simulation-based inference methods for online learning and has contributed methods that enable the mainstream use of high frequency financial data in economics. Professor Shephard joined the Harvard faculty in 2013, splitting his time equally between the Economics and Statistics Departments. He served as chair of Harvard's Department of Statistics from 2015 to 2022. In 2018, he was appointed the Frank B. Baird, Jr. Professor of Science, continuing his work in both departments. He is a fellow of several prestigious organizations including the Econometric Society, the British Academy, the Society for Financial Econometrics, and the International Association for Applied Econometrics. Prior to Harvard, Professor Shephard was a faculty member at the London School of Economics from 1988 to 1993 and at Nuffield College, Oxford from 1991 to 2013. He earned his Ph.D. from the London School of Economics in 1990.

Research topics

• Computer Science
• Machine Learning
• Artificial Intelligence
• Statistics
• Mathematics
• Econometrics
• Engineering
• Algorithm
• Mathematical optimization

Selected publications

• Deep Learning-Enhanced TopoStats for the Automated Quantification of DNA and Complex Biomolecular Structures

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-05-07

  articleOpen access

  Abstract Atomic force microscopy (AFM) enables nanometre-scale, label-free imaging of biomolecules and surfaces under near-native conditions, yet quantitative analysis of AFM data remains limited compared to other bioimaging modalities. This limitation largely arises from the absence of open, automated tools capable of addressing AFM-specific artefacts, data formats, and topographical outputs. Here, we present the latest version of TopoStats, an open-source Python package for automated and quantitative AFM image analysis, developed as a deep-learning enabled advancement of our original TopoStats software to support more complex samples and richer molecular characterisation. The pipeline integrates all key processing stages, including image flattening and noise correction, object detection and segmentation, morphometric feature extraction, and strand tracing with topological classification. Designed for accessibility and reproducibility, TopoStats adheres to the FAIR for Research Software (FAIR4RS) principles and provides configurable workflows adaptable to diverse biological samples. Combining high-resolution AFM and our analysis pipeline allows the quantification of subtle structural changes within a heterogeneous sample set, revealing properties not accessible with other structural biology techniques. We demonstrate the effectiveness of our pipeline to differentiate between plasmids with both different topology and sequence, by extracting meaningful quantitative descriptors that distinguish the samples with statistical significance. Collectively, these developments establish TopoStats as a versatile framework for high-throughput, quantitative AFM analysis, advancing AFM from a fundamentally qualitative visualisation technique toward a quantitative analytical tool.

  PublisherDOI

• Ole Eiler Barndorff-Nielsen and financial econometrics

  Bernoulli · 2025-12-11

  article1st authorCorresponding
  PublisherDOI

• How English domiciled graduate earnings vary with gender, institution attended, subject and socio-economic background (Executive Summary)

  2024-10-07 · 27 citations

  report

  This Executive Summary accompanies a new IFS working paper which uses administrative data to measure how the earnings of English graduates around 10 years into the labour market vary with gender, institution attended, subject and socioeconomic background.

  PublisherDOI

• Interactions With Sir David R. Cox

  Harvard Data Science Review · 2023-04-27 · 1 citations

  articleOpen access1st authorCorresponding
  PublisherOA PDFDOI

• Inference and forecasting for continuous-time integer-valued trawl processes

  Journal of Econometrics · 2023-07-06 · 10 citations

  articleOpen access

  This paper develops likelihood-based methods for estimation, inference, model selection, and forecasting of continuous-time integer-valued trawl processes. The full likelihood of integer-valued trawl processes is, in general, highly intractable, motivating the use of composite likelihood methods, where we consider the pairwise likelihood in lieu of the full likelihood. Maximizing the pairwise likelihood of the data yields an estimator of the parameter vector of the model, and we prove consistency and, in the short memory case, asymptotic normality of this estimator. When the underlying trawl process has long memory, the asymptotic behaviour of the estimator is more involved; we present some partial results for this case. The pairwise approach further allows us to develop probabilistic forecasting methods, which can be used to construct the predictive distribution of integer-valued time series. In a simulation study, we document the good finite sample performance of the likelihood-based estimator and the associated model selection procedure. Lastly, the methods are illustrated in an application to modelling and forecasting financial bid–ask spread data, where we find that it is beneficial to carefully model both the marginal distribution and the autocorrelation structure of the data.

  PublisherOA PDFDOI

• 1502 Geospatial visualisation of emergency department attendance rates and their associations with deprivation and non-urgent attendances

  Emergency Medicine Journal · 2022-11-22

  articleOpen access

  Aims, Objectives and Background Attendances at emergency departments in England continue to increase above the capacity of the urgent and emergency care system. There is significant variability in the rates of attendance at emergency departments across different localities. The aim of this study is to model the association of deprivation and non-urgent attendances with locality-based emergency department attendance rates. The secondary aim is to create an interactive data visualisation tool to engage stakeholders, clinicians, and the public with the research. Method and Design We undertook a retrospective, observational study using routinely collected emergency department attendance data from Yorkshire and the Humber (population 5.4 million) between January 2013 and March 2017. We calculated average annual age and sex standardised attendance rates at emergency departments for small localities known as lower layer super output areas. The association between emergency department attendance rates, deprivation and non-urgent attendances was examined using multivariable linear and logistic regression models, which were adjusted for travel time to the nearest emergency department. The data was visualised to create an interactive choropleth map using R. Abstract 1502 Figure 1 Screenshot of interactive data visualisation tool showing age and sex standardised annual emergency department attendance rates in Yorkshire and the Humber Results and Conclusion The analytical sample included 6,389,383 attendances across 2,880 localities, with a median age and sex standardised annual emergency department attendance rate of 308 per 1000 population (interquartile range 130). The fully adjusted linear regression model was significant (Adjusted R2 = 0.648, F (7, 2872)=757, p<0.001 ). Higher locality-based emergency department attendance rates were significantly predicted by each increasing decile of deprivation (β =14.8, p=0.002), each minute less of travel time to the emergency department (β = 7.86, p<0.001) and each percent greater proportion of low acuity attendances (β = 8.61, p<0.001). A large proportion of the variability in emergency department attendance rates in different geographical areas can be explained by deprivation levels and proportion of non-urgent attendances. This provides an opportunity for targeted interventions to reduce emergency department attendances.

  PublisherOA PDFDOI

• Panel experiments and dynamic causal effects: A finite population perspective

  2021 · 25 citations

  Senior authorCorresponding
  • Computer Science
  • Mathematics
  • Econometrics

  In panel experiments, we randomly assign units to different interventions, measuring their outcomes, and repeating the procedure in several periods. Using the potential outcomes framework, we define finite population dynamic causal effects that capture the relative effectiveness of alternative treatment paths. For a rich class of dynamic causal effects, we provide a nonparametric estimator that is unbiased over the randomization distribution and derive its finite population limiting distribution as either the sample size or the duration of the experiment increases. We develop two methods for inference: a conservative test for weak null hypotheses and an exact randomization test for sharp null hypotheses. We further analyze the finite population probability limit of linear fixed effects estimators. These commonly‐used estimators do not recover a causally interpretable estimand if there are dynamic causal effects and serial correlation in the assignments, highlighting the value of our proposed estimator.

  PublisherOA PDF

• Inference and forecasting for continuous-time integer-valued trawl processes and their use in financial economics

  RePEc: Research Papers in Economics · 2021-01-01

  preprintOpen access

  This paper develops likelihood-based methods for estimation, inference, model selection, and forecasting of continuous-time integer-valued trawl processes. The full likelihood of integer-valued trawl processes is, in general, highly intractable, motivating the use of composite likelihood methods, where we consider the pairwise likelihood in lieu of the full likelihood. Maximizing the pairwise likelihood of the data yields an estimator of the parameter vector of the model, and we prove consistency and asymptotic normality of this estimator. The same methods allow us to develop probabilistic forecasting methods, which can be used to construct the predictive distribution of integer-valued time series. In a simulation study, we document good finite sample performance of the likelihood-based estimator and the associated model selection procedure. Lastly, the methods are illustrated in an application to modelling and forecasting financial bid-ask spread data, where we find that it is beneficial to carefully model both the marginal distribution and the autocorrelation structure of the data. We argue that integer-valued trawl processes are especially well-suited in such situations.

  PublisherOA PDF

• Fitting Vast Dimensional Time-Varying Covariance Models

  Figshare · 2021-01-01 · 7 citations

  datasetOpen access

  Estimation of time-varying covariances is a key input in risk management and asset allocation. ARCH-type multivariate models are used widely for this purpose. Estimation of such models is computationally costly and parameter estimates are meaningfully biased when applied to a moderately large number of assets. Here, we propose a novel estimation approach that suffers from neither of these issues, even when the number of assets is in the hundreds. The theory of this new method is developed in some detail. The performance of the proposed method is investigated using extensive simulation studies and empirical examples. Supplementary materials for this article are available online.

  PublisherDOI

• Inference and forecasting for continuous-time integer-valued trawl processes

  arXiv (Cornell University) · 2021-07-08 · 2 citations

  preprintOpen access

  This paper develops likelihood-based methods for estimation, inference, model selection, and forecasting of continuous-time integer-valued trawl processes. The full likelihood of integer-valued trawl processes is, in general, highly intractable, motivating the use of composite likelihood methods, where we consider the pairwise likelihood in lieu of the full likelihood. Maximizing the pairwise likelihood of the data yields an estimator of the parameter vector of the model, and we prove consistency and, in the short memory case, asymptotic normality of this estimator. When the underlying trawl process has long memory, the asymptotic behaviour of the estimator is more involved; we present some partial results for this case. The pairwise approach further allows us to develop probabilistic forecasting methods, which can be used to construct the predictive distribution of integer-valued time series. In a simulation study, we document the good finite sample performance of the likelihood-based estimator and the associated model selection procedure. Lastly, the methods are illustrated in an application to modelling and forecasting financial bid-ask spread data, where we find that it is beneficial to carefully model both the marginal distribution and the autocorrelation structure of the data.

  PublisherOA PDFDOI

Frequent coauthors

• Ole E. Barndorff‐Nielsen

  93 shared

• Siem Jan Koopman

  Tinbergen Institute

  24 shared

• Peter Reinhard Hansen

  AstraZeneca (Brazil)

  21 shared

• Asger Lunde

  University of North Carolina at Chapel Hill

  21 shared

• Kevin Sheppard

  20 shared

• Siddhartha Chib

  Washington University in St. Louis

  20 shared

• Andrew Harvey

  19 shared

• M. Pitt

  University of Kansas

  14 shared

Education

• Ph.D., Economics

  Harvard University

  1992

• B.A., Economics

  University of California, Berkeley

  1987

Awards & honors

• Fellow of the Econometric Society
• Fellow of the British Academy