Explore our knowledge

Mads Hebsgaard

PhD fellow

Subjects
Mathematics Statistics Finance Big data Machine learning Quantitative methods

Primary research areas

High-Di­men­sion­al Co­v­ari­ance Es­tim­a­tion
Es­tim­a­tion and reg­u­lar­isa­tion of large co­v­ari­ance and pre­ci­sion matrices, in­clud­ing shrink­age meth­ods, graph­ic­al mod­els and fra­gil­ity-ad­jus­ted ap­proaches for fin­an­cial ap­plic­a­tions.
De­pend­ence in Fin­an­cial Time Series
Mod­el­ling and test­ing co-move­ments in re­turns across as­sets, sec­tors and time us­ing factor mod­els, sparse pre­ci­sion matrices, res­ampling and ma­chine learn­ing meth­ods.

I de­vel­op meth­ods for de­pend­ence in fin­an­cial time series

I study how returns co-move across assets and over time, and how to estimate these links reliably from noisy, high-dimensional financial data. Better models of dependence help investors, pension funds, hedge funds and regulators see where risk is concentrated, how it can spill across sectors and markets, and how to construct portfolios more robustly.
Methodologically, my work focuses on high-dimensional covariance and precision matrix estimation, fragility-adjusted shrinkage, and tests of sector structure in equity portfolios. I use tools from statistics, time-series analysis, resampling and machine learning to build methods that are both theoretically grounded and practically implementable.
I am motivated by problems where technical improvements in estimation can have large ex-post effects, such as more stable portfolios, clearer communication of uncertainty and a more resilient financial system. I particularly enjoy programming, quantitative finance and numerical methods.

Recent research projects

Fra­gil­ity-Ad­jus­ted Co­v­ari­ance Shrink­age

Us­ing ele­ment-wise un­cer­tainty, res­ampling and ma­chine learn­ing to im­prove high-di­men­sion­al co­v­ari­ance es­tim­a­tion

Sec­tor Struc­ture in High-Di­men­sion­al Port­fo­li­os

Test­ing wheth­er sec­tor-based risk mod­el as­sump­tions match ob­served equity co-move­ments by com­par­ing sec­tor-blocked and un­res­tric­ted de­pend­ence struc­tures

Links

Se­lec­ted cod­ing pro­jects: Open-source pro­jects on factor in­vest­ing, volat­il­ity mod­el­ling, and a game

Volpy – Ana­lys­ing the volat­il­ity premi­um
Py­thon and Jupy­ter note­books to con­struct vari­ance swap rates and ana­lyse volat­il­ity risk premi­um strategies glob­ally across as­set classes, time peri­ods, and volat­il­ity re­gimes.
‘volpy’ repo
Bet­ting Against Beta – CRSP Back­test­er
C++ pro­gram us­ing CRSP data to clean re­turns and mar­ket caps and backtest Bet­ting Against Beta for US equit­ies.
‘Betting Against Beta’ repo
Wiz­ard­Wars – C# Game Pro­ject
C# game pro­ject ex­plor­ing ob­ject-ori­ented design, simple game mech­an­ics, and learn­ing a new lan­guage for fun. Play­able.
‘WizardWars’ repo

Outside activities

Majority owner, TactileSnouts ApS, 2023–present

De­vel­op­ing a product to stim­u­late pigs (PCT pat­ent pending). Pre-rev­en­ue.