I have an unbalanced panel of around 1 million firms over 20 years. Each firm has one of the 3 status in each year, 1: active, 2: exiting (in the next year) and 3: going bankrupt (in the next year). The bankrupt firms is only about 1% of the sample while exit rate is about 10%. I am using firms' lagged values of financial ratios (such as debt ratio, current ratio, etc.) to predict their status. I would like to use the LR test shown after the xtmlogit,re model to test whether a pooled mlogit is valid.
Currently this model is running for more than 10 hours and still is in its second iteration. Since the fe model removes the firms that do not change status over the sample period, it gives me outputs faster. Are there any techniques that I can use to increase the speed of my xtmlogit,re model? If not, is there another way to test the validity of a pooled mlogit regression?
Code:
xtmlogit status debt_ratio current_ratio, baseoutcome(1) iterate(100)
