Statistics/Finance seminar with Søren Wengel Mogensen

The Department of Finance and Center for Statistics are proud to announce the upcoming seminar with Søren Wengel Mogensen, Lund University.

Søren Wengel Mogensen will present: Estimation of causal effects using covariate adjustment in vector autoregressive processes

Abstract:

In a causal linear model with a finite number of variables, linear regression can be used to estimate total causal effects when the variables included in the linear regression, the adjustment set, satisfy a set of assumptions concerning the causal graph. We extend adjustment theory to vector autoregressive processes (VAR-processes) to obtain analogous results. In our VAR-process causal graphs, each node represents a coordinate process, not a single random variable. This means that we do not need to know the order of the VAR-process, and it could even be of infinite order.

We define the total causal effect of process A on process B as a sequence of lag-specific causal effects, and we give sufficient graphical conditions for a set of coordinate processes to be an adjustment set for a total causal effect. When these conditions are met, fitting a VAR-model to the observed data will be consistent for the total causal effect using the adjustment set as covariate processes in the VAR-model. When the true order is unknown, a consistent estimator can be formulated as a limit of finite-order estimators. Several adjustment sets may exist, and we discuss efficiency in terms of asymptotic variance of the associated estimators.

With knowledge of the causal graph, one may use the theory described above to corroborate or invalidate a causal interpretation of parameter estimates obtained using standard VAR-modeling techniques. This is useful in fields where VAR-processes are common in applications, e.g., economics, and we give an example of an application.

Location:
Solbjerg Plads 3
2000 Frederiksberg
Room: SPs03