Consider two positively-valued time series, $\{X(t),Y(t)>0|t\geq0\}$. Now consider two transformations: $$ U(t) = Y(t) - \beta X(t),\\ V(t) = \ln{[Y(t)]} - \ln{[\beta X(t)]}, $$ with $\beta>0$ such that $\mathbb{E}[U(t)] = 0$. My interest is in the behaviour of $U(t)$ and $V(t)$, and whether or not they move together. My intuition (and rough simulation) tells me that they do, more or less, provided that $\mathbb{E}[U(t)]=0$ holds.

Is it possible to show that a given percentage change in $U(t)$ from $t$ to $t+1$ is accompanied by an approximately equal percentage change in $V(t)$ from $t$ to $t+1$? That is, $$ \frac{U(t+1)}{U(t)}-1 \approx \frac{V(t+1)}{V(t)}-1. $$


They do not behave similarly: Since you are taking logarithms on the parts, there is no bijective relationship between these two time-series. In particular, you have

$$\exp V(t) = \frac{Y(t)}{\beta X(t)} = 1+\frac{U(t)}{\beta X(t)} = 1 + \frac{\mathbb{E}(X(t))}{\mathbb{E}(Y(t))} \cdot \frac{U(t)}{X(t)},$$

which gives:

$$V(t) = \ln \Bigg( 1+ \frac{\mathbb{E}(X(t))}{\mathbb{E}(Y(t))} \cdot \frac{U(t)}{X(t)} \Bigg).$$

There is no way to disentangle the functions $U$ and $X$ in this equation without adding an equivalent nuisance function measuring the separate parts. As you can see, the time-series $V$ is effectively measuring ratios of the parts rather than differences of the parts, and so they can only be related if you specify one of the individual parts (or a mathematical equivalent).


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.