Modeling exchange rates

I avoided this topic for some time because I could see no way that exchange rates were anything but a very noisy channel.  This paper,  from a Thoma post, pioneered some related results.  The authors, Johansen and Juseliu did the following (from my five minute read of the paper):

They found that purchase parity differences are adjusted and these in turn adjust the internal interest rates, these the interest rate differentials alter the exchange rate as a second, relatively independent process. The two processes are separated because the agent just assumes imperfect information to reduce transaction costs.
But how did they get the two processes separated? The modeled the long swings, the secondary process, as a series of linear, discontinuous trends. And they made sure the trends chosen introduced minimal variations within measurement error. Then took the trends out and found the remaining process obeyed PPP adjustments. (I did the quick scan)
They did time dependent channel theory. Next time they will do time independent analysis and find that given the channel noise there is an optimum two partition of exchange trades. Economists do their homework.

What would be the simple undergraduate method of modeling exchange rates?
Get the original sequence of trades, if you can, or scale the time series by volume, so as to eliminate any time smoothing done by the brokers.  Then take the sequence and run them through the Windowed Huffman encoder and watch the encoding tree, just like we did with bankers.   But it will be a high noise channel, traders have bad entanglement across oceans.