This is a Julia repository to compute solutions for a discrecte rational inattention problem (Sims 2003, https://doi.org/10.1016/S0304-3932(03)00029-1) using the algorithm suggested by Thomas Cover (1984, https://doi.org/10.1109/TIT.1984.1056869).
To add this package, execute:
import Pkg; Pkg.add("RICoverAlg")
Alternatively, when being in the REPL, change into the package manager by typing ] and execute add RICoverAlg
using RICoverAlg
risolve(UtilTable, g, λ)
where 'UtilTable' is a payoff matrix: the rows correspond to the actions, the columns to the states, and an entry is the payoff the decision maker gets in the particular combination of state and action; 'g' is vector of prior probabilities over the states; 'λ' is the unit cost of learning.
import Pkg; Pkg.add("RICoverAlg")
using RICoverAlg
UtilTable = [ 0 1 0 1;
0 0 1 1;
1/2 1/2 1/2 1/2]
λ = 0.4
g = [0.45, 0.05, 0.05, 0.45]
result = risolve(UtilTable, g, λ)
result[1][1]
This gives 0.27679620629020496, which is the unconditional probability of chosing action 1.
- Here you find this code in a Pluto notebook.
- You can also run the notebook on binder
The code below creates Figure 1 of Matejka and McKay (2015, http://dx.doi.org/10.1257/aer.20130047):
import Pkg; Pkg.add("RICoverAlg")
using RICoverAlg
UtilTable = [ 0 1 0 1;
0 0 1 1;
1/2 1/2 1/2 1/2]
λ = 0.4
ρs = -1:0.1:1 res = []
results = [risolve(UtilTable, [1+ρ; 1-ρ; 1-ρ; 1+ρ]./4, λ)[1][1] for ρ in ρs]
using Plots
plot(-1:.1:1, res, label="λ=0.4", c=:purple, marker=:square, alpha=0.6,
xlabel="ρ", ylabel="Unconditional Choice Probability")
plot!(-1:.1:1, 0.5:-0.0125:0.25, label="λ=0", c=:black, marker=:o, alpha=0.6)
- Here you find this code in a Pluto notebook.
- You can also run the notebook on binder