Thursday, 12 January 2017


Then I was at universtity, my favourite text books were the "Schaum outline series". They took you through a subject with worked examples, so you could see how it was done, and tons more examples so you could learn-by-doing.

I've been reading popular books on quantum mechanics, and I suddenly realised that I should get the Schaum book. So I went on Amazon and ordered it, about £5. Just so you know how cheap that is, 50 years ago, the typical textbook cost £25.

So while I was browsing, I saw the Schaum books on econometrics and on mathematical finance, and I thought about when I made the jump from maths to econometrics.

The big thing I discovered, which I don't think is widely known, is how inaccurate economic statistics are.  They are so inaccurate, that they really cannot support the weight of analysis that is put on them.

GDP figures, for example, are collected via a survey of a sample of companies. Consumer expenditure is measured via the "Family income and expenditure survey", with, as I recall, a sample of 7000.

And my best example of the problems with data, comes from population, which you'd think would be *very* accurate, since it's collected by a 100% census.

Now look at the population of Germany. If you want to use that in a time series analysis, there's a huge "gotcha". In 1964 (I might have misremembered the exact date) there's a jump of two million in the time series. But what isn't obvious, is that there weren't suddenly two million more Germans. The two million is because they started to include West Berlin in the numbers.

Before you do any analysis on data, you ought to understand how it was collected, and in what ways it's inaccurate. But I don't think economists ever do that.

No comments:

Post a Comment