The equity markets have been weak on the sovereign risk scares in Greece, Spain and Portugal. As much as this, employment numbers in the US have now come in below forecast in the last 4 weeks.
People out of work tend to stop their efforts in December as the year comes to a close because they know employers stop their searches close to the year end. In the new year, the number of people who return to actively seeking work picks up. This explains why the initial jobless claims are usually in bigger deficit every January, February.
Macro data are often seasonally adjusted to take into account temporal dependencies in the data.
Similarly, forecasts of macro data often use the same estimation models to arrive at their forecasts. Say that we have a fairly comprehensive model that assumes that the data exhibits seasonality, cyclicality, momentum, mean reversion (more generally than suggested by cyclicality and seasonality), a general drift. Models are estimated based on some assumption as to the stability of the uncertain part of the data. It is always assumed that the data consists of a portion that can be explained (by explanatory variables, by autocorrelations, moving averages etc.) and a portion that cannot. Before a model can be estimated and ready for use in forecasting, assumptions need to be made about the behaviour of the portion that cannot be explained.
Unfortunately, when you have a big crisis like 1930, 1970, 1987, 2001, 2008, the data violates most of the assumptions in previously stable models. There is no good way of dealing with extreme events which by definition don’t happen very often. You just can’t calibrate a model on a few pieces of data. In any case, the inclusion of crisis period data simply confounds models and their predictive abilities.
Like now. Models calibrated on pre crisis data are no good in the crisis. The same models calibrated on crisis data, are simply no good for post crisis prediction. It is far more useful to understand the dynamics behind the numbers. The way in which initial jobless claims are calculated introduce lags, introduce behavioural phenomena which are hard to quantify. Expect more volatility in the series as the number of people who come back to the job market inflate this unemployment measure. As the economy recovers, expect to see the measure lag, and overshoot on the recovery.