Initial Claims, yet again

As I expect this interests me more than anyone else, I'll try to restrain myself a bit here, but this line is the long-run (two weeks out) prediction that would be made based on data up to the time for which it's plotted using an ARIMA(0,1,2) model with both MA coefficients set to -0.2. It clearly hugs the data fairly well over any period of more than a couple weeks, it's previsible at the time of the data release to which you would compare it, and the change from one period to the next exhibits one third of the variance of the original series.

The last point on the new curve is 555,000, down from 590,000 the week before; the last two points have been significantly below recent values, and this figure responds more quickly to repeated deviations like that than the exponential moving average I posted before, while still filtering out two-thirds of the noise. I might, however, persist in using the exponential moving average just because 1) it's simpler to explain, and 2) it requires less "state"; if I give you today's exponential moving average, and you see next week's release, it's easy to calculate next week's average, while this requires two data points and is less mathematically simple. I haven't decided yet which I prefer, and I'd welcome any input anyone else wants to offer.

Labels: Initial Claims of Unemployment

- posted by dWj @ 1:47 AM