Locations of example data that are included in the package for convenience.
Location of a CSV file containing the fake cereal product data from Nevo (2000). The file includes the same pre-computed excluded instruments used in the original paper.
Location of a CSV file containing the fake cereal agent data. Included in the file are Monte Carlo weights and draws along with demographics, which are used by Nevo (2000) to solve the fake cereal problem.
Location of a CSV file containing the automobile product data extracted by Andrews, Gentzkow, and Shapiro (2017) from the original GAUSS code for Berry, Levinsohn, and Pakes (1999), which is commonly assumed to be the same data used in Berry, Levinsohn, and Pakes (1995).
The file also includes a set of optimal excluded instruments computed in the spirit of Chamberlain (1987) for the automobile problem from Berry, Levinsohn, and Pakes (1995), which are used to solve the problem in the tutorial. These instruments were computed according to the following procedure:
Traditional excluded BLP instruments from the original paper were computed with
build_blp_instruments(). As in the original paper, the
mpdvariable was added to the set of excluded supply-side instruments.
Each set of excluded instruments was interacted up to the third degree, standardized, replaced with the minimum set of principal components that explained at least 99% of the variance, and standardized again.
These two sets of principal components were used as excluded demand- and supply-side instruments when solving the first GMM stage of a
Problemconfigured as in the tutorial, but with non-optimal instruments.
compute_optimal_instruments()method was used to estimate the optimal excluded instruments for the problem, which were standardized.
Location of a CSV file containing automobile agent data. Included in the file are 200 Monte Carlo weights and draws for each market, which, unlike in the fake cereal data, are not the same draws used in the original paper.
Also included is an income demographic, which consists of draws from lognormal distributions with common standard deviation
1.72and the following market-varying means:
These numbers were extracted also extracted from the original GAUSS code for Berry, Levinsohn, and Pakes (1999).