Hi QC Community
 
We're pleased to launch several demo algorithms on alternative data usage to stimulate some thoughts on how to use them for trading. In this instance, we'll be using Brain Language Metrics on Company Filings dataset. This dataset implemented natural language processing in 10-K and 10-Q reports for over 3000 US equities to generate sentiment levels on various categories in these reports.
 
The idea behind this algorithm is that it is commonly observed that stock price would experienced large fluctuation after financial reports published metrics that were out-of-expectation, as the forecast of the company's future changes. This brings up the idea that the volatility will become larger if the sentiment is extra high in magnitude regardless of positive or negative.
 
In this algorithm, we'll trade the volatility mean reversion by short Iron Condor, as well as collect time decay value. We'll be using an ARIMAX model to model the volatility with our Brain's dataset as exogenous variables (just like GARCH), and then use SDE and Monte Carlo simulation to simulate the price series process by the predicted volatility. The predicted volatilities will be used as entry signals and the price simulations will be used as strike price decisions on the options we used for the Iron Condor.
 
The attached notebook has included the development and validation of the hypothesis, and how we translate the results into developing the strategy and algorithm of trading volatility via Options using iron condor. Please find and follow the detailed explanation in the research notebook then the backtest. Enjoy!
 
Best
Louis Szeto

Author