forecasting exchange rates

[flyingfox2002]

If you have a small sample of real foreign currency exchange rates (e.g. Yen to Dollar), how would you use regression to forecast what the exchange rate will be at... say, 3 months from today?

My workgroup proposed using the same real data as both X variable and Y variable, using the data as X, then simply skip ahead 3 months, and use the same as Y.

For example, lets say we have data for Sept05, Oct05, Nov05, Dec05, Jan06 and so forth until May07.

Sept05's data would be X1, Oct05's would be X2 and so forth. We then would skip 3 months ahead and use Dec05's data as Y1, Jan06 as Y2, and so forth. Then we plot the data in Excel and use it to find a Regression equation.

The prof shot it down though so we are stumped.

Any pros here have any suggestion?

Thanks

[darkcser]

If you have a small sample of real foreign currency exchange rates (e.g. Yen to Dollar), how would you use regression to forecast what the exchange rate will be at... say, 3 months from today?

My workgroup proposed using the same real data as both X variable and Y variable, using the data as X, then simply skip ahead 3 months, and use the same as Y.

For example, lets say we have data for Sept05, Oct05, Nov05, Dec05, Jan06 and so forth until May07.

Sept05's data would be X1, Oct05's would be X2 and so forth. We then would skip 3 months ahead and use Dec05's data as Y1, Jan06 as Y2, and so forth. Then we plot the data in Excel and use it to find a Regression equation.

The prof shot it down though so we are stumped.

Any pros here have any suggestion?

Thanks

Not a pro but I'll take a stab in the dark at it.

The exchange rates are given as a fraction of some sort 110yen/1$ or some such right?

Plot the exchange rate as Y values with the month as X values.

Find the regression line y=mx+b either through Excel or using least squares.

Now plug in the month you want to forecast for into the equation as x then you should get the exchange rate.

note: You should treat the month as month1, month2, etc. instead of Dec 05. This is assumed to be a linear regression.

[warlock110]

Let see if i can remember this. If i remember correctly, regression has different types, the easier one being linear regression, and the more complicated ones are having to do with least square (accounting how far each of the data is from the "line", i think this type appears in econometric/stats/linear algebra @ calculus level). I'm not sure but i believe exel can do both, i've never used it, i know stata and some other stat software can do both, and even the TI-83 has functions in there to do it.

The basic concept is to sort of plot all the data points that you have, and find the "line of best fit", taking that line (some sort of equation, whether it's a linear line or curve, you should still have some sort of equation for the line). You take the point that you want to "predict", and plug into that equation. Exel should be able to do it, but if you ask me how, i don't know. I can do it by hand or using calculator/stata (i hope haha), but i've never used exel for something like this before.

[flyingfox2002]

That's what my group partner did hehe. Month as X, data as Y. It worked. It didn't occur to me that month could work as a variable. (But the forecast isn't so great because it ended up predicting a steep drop of the exchange rate into the future... but ah well we don't care )