Submission: Tingjun attempt 2

[tingjun-yang]

Hi David,

I would like to withdraw my previous submission (tingjun) and replace it with a new GARCH(1,1) Volatility Model (labeled tingjun2). My original model had some structural issues that I believe are resolved in this new iteration.

This GARCH(1,1) implementation utilizes three optimized parameters that yield a highly stable q-variance fit. On a 100,000-day simulation, the model recovered fit parameters of $\sigma_0 = 0.252$ and $z_{off} = 0.026$, which align closely with the target values of $0.259$ and $0.021$. Crucially, the $R^{2}$ for this fit is 0.996, exceeding the 0.995 entry threshold.

While testing on a 5M-day sample showed a slight divergence in fit—likely due to the sensitivity of global de-meaning to sample size—I am confident that re-tuning the parameters for larger datasets will maintain this high level of performance.

I have attached the updated model documentation and results and a csv file with 100k prices (simulated_prices.csv). I look forward to your feedback.

Best regards,
Tingjun Yang

[Orrell]

Hi @tingjun-yang Thanks for this, we will take a look. One thing is that as mentioned in the README the aim is to fit the specific parabola shown there in Figure 1 (otherwise people could ignore the drift term, and it would not be a level playing field). For consistency, can you revise your parameters and rescore? I commented out a line in the score_submission.py file to make this clearer. Best, David

[Orrell]

Hi @tingjun-yang Please do that test using the full length dataset, not just the shorter simulations. Also please declare the 0.01 in the line V = omega + (lam * V) + ((1 - lam - 0.01) * ret_sq) as a parameter since it affects the results. Thanks, David