Intuitive Question

Suppose I'm given a set of $k$ time-series $\{X_t^1,\dots, X_t^k\}$. Is there a way to determine how much of each series is dependent on the others.

Formal Question

More precisely if for each $j \in \{1,\ldots,k\}$, $X^j_t$ is a stochastic process with values in $\mathbb{R}$, $\varepsilon_t^j$ is a random variable in $\mathbb{R}$ and define the loss function: $$ \mathfrak{L}_j(a_1,\ldots,a_j):= \int \left( 1-\sum_{i\neq j} a_i \right) \sum_{i \neq j} (a_i X^i_t - X_t^j)^2 \, dt $$

Then how can we determine the best $a_i$ minimizing: $$ X_t^j = \sum_{i \neq j} a_i X_t^i + a_j\varepsilon_t^j, $$ where $\{a^1,\ldots,a^j\}$ are non-negative real numbers such that $$ \sum_i a_i =1 $$ is a convex combination; how can we find the best $a_i$; that is how can we determine the $a_i$ minimizing $\mathfrak{L}(a_1,\dots,a_k)^j$?.

If it makes things simpler we may assume that the processes are time-series...


Just estimate regression equations, e.g. $$X_1=\theta_0+\theta_1X_2+...+\theta X_k+\epsilon.$$ Then change dependent variable and estimate $$X_2=\phi_0+\phi_1X_1+...+\phi X_k+\mu.$$ And so on.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.