The RATS Software Forum

Dear Tom Doan

I am writing again for your great help. I believe that you will forward your helping hand again. I am going to submit my PhD thesis shortly. I am facing the problem is that when I am doing simple regression between Australian dollar exchange rate and Australian steam coal export. The result shows that 1 cent increase Australian dollar value against US dollar, Australian steam coal export decreases by 77,230 tonnes. Please note that Australian dollar exchange rate and Australian steam coal export are non-stationary at level but 1st difference, they are stationary and they are not co-integrated at level. Therefore, I consider 1st difference data for simple regression analysis. On the other hand, when I am doing multiple regressions considering Australian steam coal export as a dependent variable and independent variables are Australian dollar exchange rate, Australian steam coal price, Australian labour cost, lending interest rate, GDP growth of Japan, Korea, India China and Taiwan. They are all non-stationary at level but 1st difference they are stationary. I also find that they are all co-integrated. So, I consider the data at level. After finishing the analysis, I find that 1 cent increase the value of Australian dollar with US dollar, Australian steam coal export increases by 115,800 tonnes. The problem is that 1 cent value increase of Australian dollar, the Australian steam coal export decreases by 77,230 tonnes in simple regression but in multiple regression increases by 115,800 tonnes. Could you please explain why it was happened and which result is correct and how do I explain it?

I am looking your best help and co-operation.

Thanking you
Ali

Why would you think that regressing exports on the exchange rate (only) would give any useful result? If you've shown that the two are I(1) and not cointegrated, you're running exactly what the "spurious regression" result warns about. This is not saying that the bigger regression is correct, but the small one seems clearly incorrect.

Dear Tom Doan

Thank you for your reply and help. Could you please explain little bit more your answer for me? Because I am not fully clear your answer and how can I overcome this problem?

I am looking your best help and co-operation.

Thanking you

With Kinds

Ali

If you tell me that Y is I(1), X1 is I(1) and Y on X1 is I(1), but that Y on X1, X2, X3, ... is I(0), then it's rather clear, without even looking at the actual output, that the sum of squared residuals is much lower in the second case than the first. In reply to almost every post that you've made, I've urged you to think about the economics of the situation and not just the I-ness of the series involved. Why would you think the prime determinant of Australian coal exports would be the Aus/US $ exchange rate? Doesn't Australia ship coal to Asia, not the U.S? If you're looking at an exchange rate, wouldn't the Aus $ vs the yen, or some basket of Asian currencies make more sense? Also, why would you even suspect that there's a static relationship? How long does it take to ship coal? How long in advance are contracts in the coal market? Aren't you going to need a well thought-out dynamic model rather than a static one?

Dear Tom Doan

Thank you for your replay and explanation. Australian coal is sold in US dollar invoice. Therefore, I am trying to measure cause and effect relationships between exchange rate A$/US$ and Australian coal export, production and pricing by simple regression. I also like to check how are the key factors (such as exchange rate, GDP growth of Australian coal importing countries, labour cost of Australia, lending interest rate of Australia etc.) influencing the Australian coal export? I am completing by multiple regression and the results shows that the sum of squared residuals is much higher in the second case than the first. I did '"Spurious Regression test, Wald test and Heteroskedascity Test". All these test support my result.The only problem is that 1 cent value increase of Australian dollar, the Australian steam coal export decreases by 77,230 tonnes in simple regression but in multiple regression increases by 115,800 tonnes. Logically it does not support. if you kindly inform me how can I defend it.

I am looking best help in this regard.

Thanking you

With Kind
Ali

How can you possibly run a multiple regression with added variables and have the sum of squares higher with added variables than without?

If the contract price is in US$, wouldn't a decline in the US$ increase demand? Do you think you're estimating a supply function? A demand function? The typical combination of the two that you learn about when you do simultaneous equations? Given the timing of contracts, would you even expect that current changes in exchange rates would have any effect at all?

You're asking questions that you should be addressing to your advisor.