“Sometimes, a little knowledge is a dangerous thing, and that goes double in finance. The client thought he knew what he was asking for, and he thought he knew what needed to be done. If only it were true!
Of course, I bear some responsibility for all of this, as I had ample opportunity to realize that my clients lack of understanding was likely to lead to problems down the road.
My first clue should have been that the initial description of the project, something to simply find trends in the markets is not simple. If it were, we would all be rich!
But it turned out that he really wanted a determination of which of the stock market indices of Brazil, Russia, India, and China was least correlated with those of the first world. Now anybody who knows anything about this would know that there are several gotchas involved in getting quantitative answers here. Anybody, that is, except my client.
My next clue that I was working with an ignorant and difficult client was that the conversion of the closing index quotes from local currency to the relevant first world currency was not in the job specification. Nor was there any mention of statistical tests to see if correlation meant anything in the context of interest (The point of the clients project was, presumably, to make a decision regarding future action on historical analysis. However, if the history shows that things are constantly changing, then there is no reason to think that any specific historical conclusion will be relevant!).
Another gotcha in all of this is the fact that world stock markets dont close at the same time. So, should todays returns from the New York Stock Exchange be correlated with yesterdays returns or todays returns on the London exchange? This is especially vexing, given the clients interest in Granger causality. Once again, the client didnt quite understand the careful work that I had to do to tease all of this out. In particular, he didnt understand that the high correlation between some of the close-to-close returns is an artifact of news arriving more or less contemporaneously at two exchanges at the same time. True Granger Causality would obtain only if there were, for example, a high correlation between the same-day Nikkei and S & P returns because these two exchanges are never open at the same time. As a result, I came to a different conclusion than the client expected, which may well be one of the real reasons for his pique.
I also concluded that the effects that the client was looking at were too small to be meaningful, even if they were statistically significant. For example, the 16% correlation found between New York and Tokyo result in an R-squared of 2.6%. Please.
At that point, well, I was just tired of the whole thing that the client didnt seem to understand what was involved, numerous demands for extra tests, and the fact that I was working on a fixed price job that, frankly, I underbid. I admit that I didnt run some of the tests that I had said that I was going to run (having run others that I thought more relevant to the task at hand and having thought the question of the statistical significance of an R-squared of 2.6% to be silly). I had hoped to work this out amicably with the client, but he placed the job in problem status. At that point, I was forced to let Elance sort it out.
Moral: If you dont know what you want and you arent willing to pay for someone to tell you what to want, you should not expect a good outcome. And, if I cant detect problem clients like this one, given the clues above, I shouldnt expect a good outcome, either.