Choosing Currency Units and Return Measures for Forex Modeling

Summary

The document raises preprocessing choices for modeling exchange-rate data. It asks whether rates quoted against different currencies should be converted to a common base currency, illustrating how a missing cross rate can be derived from two available rates. Expressing all pairs relative to one currency may make the series easier to interpret and support profit calculations.

The question also distinguishes modeling exchange-rate levels from their changes or returns, and asks whether the appropriate representation depends on the model. The document provides no answer, empirical comparison, or recommended transformation. It is therefore useful as a framing of currency-unit consistency and target-variable choices, but it does not establish that a common base currency is always preferable or prescribe a modeling method.

Key ideas

• Exchange rates quoted against different currencies can be converted to a common base using cross-rate relationships.
• A common currency denomination may make comparisons and profit calculations more interpretable.
• Modeling levels, changes, or returns are distinct choices that may depend on the model.
• The document poses these choices but provides no recommended preprocessing method or evidence.

Tags

# Common pre-processing steps for forex data


# Common pre-processing steps for forex data












This is almost certainly going to make me seem like a novice, but googling for answers is very difficult for this sort of thing.

My friend asked me to take a look at some forex data recently. My instincts for vector-valued time series are to plot some pictures and try models. If I have to pre-process the data, I generally know what's acceptable.

This is my question: is it more interpretable to convert everything into the base currency? Is this commonly done?

Say we have four countries: A,B, C and D, and I'm from country C. Say we have available at some time $(\frac{A}{D}, \frac{A}{C},\frac{B}{C})$.

Instead of modelling $(\frac{A}{D}, \frac{A}{C},\frac{B}{C})$ directly shouldn't we model $(\frac{A}{C},\frac{B}{C},\frac{D}{C})$? Here $\frac{D}{C} = \frac{A}{C}/\frac{A}{D}$.

Now everything is in the same units, and this also facilitates profit calculations. Or does all of this depend on the models I'm using? Is it more common to model these levels, or their changes? Or maybe even their returns?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.