4 Value Decomposition¶
Value decomposition is
where
There are two steps to do value decomposition:
Simulate state variable processes and first variation processes.
Calculated four terms of discounted social cash flow.
4.1 Simulate First Variational Process and State Variables¶
FeymannKacs_prepare.sh and FeymannKacs_simulate.sh are two main bash files.
FeymannKacs_prepare.py loads solutions from solved HJB equations, computes related variables in state space, calculate derivatives via finite difference and interpolate them.
FeymannKacs_simulate.py simulates the first variational process, the distorted state variables and other terms needed in value decomposition.
\[M_t= \begin{bmatrix} M_t^{ \log \tilde{ {K}}} \cr M_t^{\tilde{Y}} \cr M_t^{\log\tilde{{R}}} \cr M_t^{\log\tilde{{N}}} \end{bmatrix},\quad \tilde{X}_t= \begin{bmatrix} \log \tilde{ {K}}_t \cr \tilde{Y}_t \cr \log\tilde{{R}}_t \cr \log\tilde{{N}}_t \end{bmatrix}\]For notation clarification, we using \(\log{K}\) instead of \(\hat{K}\) in this section.
To simulate the first variational process with respect to technology, we set the initial value of \(M_t\) to be \([0,0,1,0]'\), and \(\tilde{X}_t\) to be \([log(\frac{85}{0.115}), 1.1, log(11.2),1.1 \gamma_1 + 0.5\times 1.1^2\gamma_2 ]'\). Initial values are set in code line 905.
We use \(M^{\log \tilde{ {K}}_t}\) as an example and others are the same. The implied evolution of the process \(M^{\log \tilde{ {K}}_t}\) is given by
, where \(\tilde{X}_t\) are distorted state variables.
Recall that the distorted Capital evoluation is
where \(\sigma_{\log \tilde{ {K}}_t} = \sigma_k\) and \(\mu_{\log \tilde{ {K}}_t}\)
Line 303 shows how we calculate those derivatives by finite differnece.
Line 397 interpolates the vectors from finite difference to get the functions.
Then we start the for loop from time 0 to recursively get four discounted term.
4.2 Calculate four value decomposition terms¶
Besides above variables, the derivatives we need in value decomposition are
We use finite difference and interpolation same as above. Line 313 shows how we code it. Now we have every element in value decomposition.
We also consider four different configurations of uncertainty aversion as a way to assess the different economic forces in play:
pre-jump neutrality - post-jump neutrality;
pre-jump neutrality - post-jump aversion;
pre-jump aversion - post-jump neutrality;
pre-jump aversion - post-jump aversion.
FeymannKacs_simulate_NewPlug.sh controls different \(\xi\) for different channels.