Assistants

When you add a trading idea to your Research Pipeline, the Research Assistant takes it directly. From there, every step of the analysis occurs inside the Research Environment and is documented like a research report, walking the next reader through how the data was pulled, what was tested, and what was concluded.

The Research Assistant builds a feature set around your idea by pulling in ten competitors along with commodity and macroeconomic data, then describes the series in plain terms — date range, frequency, missing observations, and the mean, variance, skewness, and kurtosis of every variable. It transforms the raw data into the shapes that actually matter for time series work, generating first differences, seasonal differences, and log transforms, then plots the originals and transformations together so patterns are visible rather than implied.

Variable selection runs twice on purpose. The assistant ranks features by Pearson, Spearman, and Kendall correlation to pick the top five features, then fits a decision tree and pulls the top five most important features from that as well. Five candidate models are fit against those features, and the assistant keeps the ones with good accuracy whose residuals pass an ADF stationarity test with no trend and no constant. If nothing survives, it widens the search to twenty variables and tries again before reporting honestly that no model fit cleanly.

A notebook that documents the data, the transformations, the variable selection process, the candidate models, the ACF and PACF charts of the residuals, the ADF test results, and a written summary of why the top models work and what assumptions sit underneath them. The Research Assistant does not backtest but by the time the notebook is finished, your team of Assistants know whether the idea is worth coding or whether it should be set down before it costs you any more time.

The Research Assistant has access to the following tools:

• create_compile
• read_project_nodes
• jupyter_create_cell
• jupyter_read_cell
• jupyter_update_cell
• jupyter_delete_cell
• jupyter_execute_cell
• jupyter_create_notebook
• jupyter_read_notebook
• jupyter_execute_notebook
• financial_data_blog_posts
• financial_data_news_articles
• financial_data_web_get
• datasets
• user_input
• environment_library_support
• object_store_get
• object_store_set