Previous Entry Share Next Entry
(no subject)
So anyway, a few other bits and pieces while I'm awake and (vaguely) thinking..

As previously mentioned, I've started my new job.. I am now a Forecasting Analyst, a title which includes the words "fork", "arse" and "anal" - I can tell you're jealous. The first order of business is to come up with a model that can accurately forecast work volumes for my old department, which is proving to be rather tricky.. From one month to the next, I've seen 50% increases in volumes without any corresponding changes in any of the key drivers I'm monitoring.

It might be that in the short term, the best model I can come up with is one that tries to identify common patterns, and matches them against current volumes. For example, if our work volumes peaked last month, and are starting to come down this month, then even though a recent average would be quite high, the model could perhaps try to continue any trends that we see, based on some patterns from the historical data we have. I'll have to see how that one pans out - for the time being, I don't have the historical data to work with (though I've written some Excel macros to get hold of it).

So yes, while a long range forecast may not be possible, it might at least be viable to estimate the patterns over the next couple of months, given where we think we are on a curve right now. My main concern is that a long range forecast will be impossible - that the key drivers behind the work volumes are outisde the scope of what we can work with. Unless we already have forecasts for those indicators, we can't build a forecast for anything that relies on them. And I've been here before...

Four years ago, when I was working at Systech, I was asked to build a model to assess the performance of sensor cells. The problem was, each one behaved differently - they peaked at different levels, had different baselines, and curved differently. While batches of cells were tested at the same time, due to temperature variations you could never replicate the same conditions, and due to lasting effects of testing on the cells themselves, each cell would behave differently if put through the process a second time anyway. In short, it was pretty much impossible to model. The problem was in how long it took me to convince everybody else that this was the case.

Hopefully with this job, because my immediate boss is the expert on forecasting, if I find it's impossible to build an accurate model, he should be able to look at the figures and either agree, or correct me.


Log in

No account? Create an account