Periods and instruments

Uncollapsed GMM-style instruments number about T² per variable. This suite holds 5,000 panels and adds periods, to 40 (about 1,600 instruments). The model is two-step difference GMM with Windmeijer-corrected standard errors.

The run was on an EC2 spot instance (c8id.4xlarge: 16 vCPUs, 32 GiB, local NVMe), with a 1,800-second limit per fit and the benchmarks’ memory capped at 96% of it. Each fit is in its own process.

• dpd-stream never builds a panel’s instrument matrix. Each row adds its few nonzero instruments into the moment sums, so time grows with the instruments per row and memory with the j × j sums. 40 periods take 6 seconds and 1.6 GB.

• The others hold each panel’s dense instrument matrix. At 20 periods plm needs 9 GB and 47 seconds, pdynmc 18 GB and 441 seconds, and pydynpd 24 seconds. At 30 periods plm and pdynmc run out of memory; pydynpd, which stays under 3 GB, takes 112 seconds, and 419 seconds at 40 periods.

The numbers are in benchmarks/results/instruments.csv.