This repo contains the framework for obtaining optimal trading strategies among multiple agents while finding stock market equilibrium. The framework is implemented using the generative adversarial infrastructure with deep neural networks discretized by time steps.
The special case with following assumptions is considered:
- the dynamic of the market satisfies that return
and voalatility
are unknown;
- the cost parameter
for each trading agent is constant, but different agents might have different
- the endowment volatility is in the form of
where
is constant for each trading agent. Again, different trading agents might have different endowment volatilities;
- the frictionless strategy satisfies that
and
The general variables and the market parameters and their corresponding values in the code are summarized below:
| Variable | Value | Meaning |
|---|---|---|
q |
2 | power of the trading cost, q |
S_OUTSTANDING |
1 | total shares in the market, s |
TIME |
1 | trading horizon, T |
TIME_STEP |
252 | time discretization, N |
DT |
1/252 | |
GAMMA |
risk aversion, |
|
XI |
endowment volatility parameter, |
|
PHI_INITIAL |
- | initial holding, |
ALPHA |
1 | market volatility, |
MU_BAR |
- | market return, |
LAM |
0.1 | trading cost parameter, |
test_samples |
300 | number of test sample path, batch_size |