BEGIN:VCALENDAR VERSION:2.0 PRODID:-//AMSE//Event Calendar//FR CALSCALE:GREGORIAN METHOD:PUBLISH BEGIN:VEVENT UID:event-12532@amse-aixmarseille.fr DTSTAMP:20260414T142422Z CREATED:20260414T142422Z LAST-MODIFIED:20260414T142422Z STATUS:CONFIRMED SEQUENCE:0 SUMMARY:phd seminar - Maha Ouali*\, Simon Rebeyrolles** DTSTART:20260217T100000Z DTEND:20260217T113000Z DESCRIPTION:*Accurately estimating treatment effects in time series is esse ntial for evaluating interventions in real-world applications\, especially when treatment assignment is biased by unobserved factors. In many practica l settings\, interventions are adopted at different times across individual s\, leading to staggered treatment exposure and heterogeneous pre-treatment histories. In such cases\, aggregating outcome trajectories across treated units is ill-defined\, making individual treatment effect (ITE) estimation a prerequisite for reliable causal inference. We therefore study the probl em of estimating the average treatment effect for the treated (ATT) by firs t recovering individual-level counterfactuals. We introduce a neural framew ork that learns simultaneously low-dimensional latent representations of in dividual time series and propensity scores. These estimates are then used t o approximate the individual treatment effects through a flexible matching procedure that avoids classical convexity constraints commonly used in synt hetic control methods. By operating at the individual level\, our approach naturally accommodates staggered interventions and improves counterfactual estimation under latent bias\, without relying on explicit temporal modelin g assumptions. We illustrate our approach on both real-world energy consump tion data and clinical time series\, including high-frequency electricity d emand-response programs and semi-synthetic data for individuals in intensiv e care unit (ICU)\, where hidden confounding\, staggered treatment adoption \, and non-stationary dynamics are prevalent.**This paper provides new evid ence on the effects of access to public housing in France on a large set of variables\, such as housing quality\, neighborhood characteristics\, and l abor outcomes. To this end\, a unique and novel linkage between data on soc ial housing applicants and tax records is implemented. The empirical strate gy exploits the fact that\, in areas where the imbalance between the supply and demand for social housing is substantial\, allocations can occur rando mly among applicants with similar observable characteristics. To approximat e this quasi-experimental setting\, I combine propensity score matching bet ween recipients and non-recipients with a staggered difference-in-differenc es design. The results suggest that recipients of social housing experience a trade-off in their living conditions. In particular\, their housing qual ity improves substantially\, notably through reductions in overcrowding and rent burdens. However\, the likelihood of residing in a disadvantaged neig hborhood increases following allocation. Although the theoretical literatur e predicts deteriorated labor market outcomes due to a disincentive channel or spatial mismatch\, no significant impact is observed on recipients’ e mployment trajectories in the short run\, regardless of the dimensions of h eterogeneity considered.\\n\\nContact: Xavier Chatron-Colliet: xavier.chatr on-colliet[at]univ-amu.frArmand Rigotti: armand.rigotti[at]univ-amu.fr\n\n Plus d'informations: https://amse-aixmarseille.fr/en/events/maha-ouali-simo n-rebeyrolles-0 LOCATION:Îlot Bernard du Bois - Amphithéâtre\, AMU - AMSE\, 5-9 boulevar d Maurice Bourdet\, 13001 Marseille URL;VALUE=URI:https://amse-aixmarseille.fr/en/events/maha-ouali-simon-rebeyrolles-0 CONTACT:Xavier Chatron-Colliet: xavier.chatron-colliet[at]univ-amu.frArmand Rigotti: \;armand.rigotti[at]univ-amu.fr TRANSP:OPAQUE END:VEVENT END:VCALENDAR