I’m trying to find Julia packages to perform (financial) quantitative analysis under uncertainty, typicaly by Monte Carlo simulation. Somewhat similar to Oracle Crystal Ball or Palisade @RISK. Found MCHammer.jl that seems great. However the last commit was almost two years ago and the documentation link appears to be broken. I am wondering if alternatively, Turing, Gen or some other probabilistic programming package could be easily adapted for this purpose. Any suggestion would be much appreciated!
1 Like
Not sure what these libraries are doing exactly, but from a quick look on Crystal Ball it seems to do little more than propagation parameter uncertainties forward. To this end, I would start with MonteCarloMeasurements for representing uncertain numeric quantities:
using MonteCarloMeasurements
function npv(cashflow, interest_rate)
d = 1 / (1 + interest_rate)
sum(v * d^t for (t, v) in pairs(cashflow))
end
julia> r = 0.02 ± 0.003 # 2% ± 30 basis points
0.02 ± 0.003 Particles{Float64, 2000}
julia> cashflow = Dict(0:3 .=> [-2, 1, 1, 1]);
julia> npv(cashflow, r)
0.883966 ± 0.0169 Particles{Float64, 2000}
In case, you need to simulate more complicated stochastic model DifferentialEquations.jl has stochastic differential equations. For fitting/calibrating models, Turing.jl or similar libraries would be a good start.
It might well be that libraries more specific to financial modeling exist, I’m just not aware of them.
3 Likes
While primarily known for Bayesian inference, I am wondering if Turing.jl can also be used for uncertainty propagation through Bayesian methods?
Turing is a tool for solving the inverse problem, i.e., given some output, infer the distribution of the input. This is a much harder problem in general, and typically makes use of a much wider set of tools, e.g., automatic differentiation and HMC.
1 Like