Quantitative Research with Cortex Code and AI Functions using Snowflake Public Data

Transform unstructured earnings call transcripts into actionable investment insights using Snowflake Cortex AI, ML Model Registry, and intelligent agents - all accelerated by Cortex Code.

Why This Matters

Financial analysts spend countless hours manually reviewing earnings call transcripts. This guide demonstrates how to systematically process unstructured data at scale using AI Functions (AI_COMPLETE, AI_SQL) - turning raw transcript text into structured sentiment scores, analyst participation metrics, and investment signals that feed directly into quantitative models.

Full Guide: For detailed architecture, business impact, and use cases, see the Snowflake Developers Guide.

What You Will Learn

• How to use Cortex Code to build data pipelines through natural language
• How to extract structured insights from unstructured text using AI_COMPLETE()
• How to train and register ML models in Snowflake's Model Registry
• How to create semantic search over unstructured data with Cortex Search
• How to build a Cortex Agent that combines multiple AI tools

What You Will Build

• A sentiment analysis pipeline that scores earnings call transcripts (1-10 scale)
• A LightGBM stock return prediction model with walk-forward validation
• A Cortex Search service for natural language queries over sentiment data
• A Cortex Agent accessible via Snowflake Intelligence

Prerequisites

• Snowflake account (sign up for a free trial) with ACCOUNTADMIN access (see note below)
• Access to Snowflake Marketplace

Note on Privileges: This guide uses ACCOUNTADMIN for simplicity in demo and learning environments. For production deployments, follow the principle of least privilege by creating a dedicated role with only the specific grants required.

Getting Started

Step 1: Get Marketplace Data

1. Navigate to Snowflake Public Data (Free) in Snowflake Marketplace
2. Click Get and accept the terms
3. Keep the default database name SNOWFLAKE_PUBLIC_DATA_FREE
4. Grant access to ACCOUNTADMIN role
5. Click Get to install

Step 2: Run Setup Script

1. In Snowsight, navigate to Projects > Workspaces
2. Create a new SQL file and copy the contents from scripts/setup.sql
3. Run the entire script

This creates the complete demo environment including database, warehouse, tables, stored procedures, and deploys notebooks from this repository.

Step 3: Run Notebook - Start Here

1. In Snowsight, navigate to Projects > Notebooks
2. Switch your role to FSI_DEMO_ROLE (bottom-left, click on your username)
3. Open the START_HERE notebook
4. Run all cells to extract analyst sentiment from earnings call transcripts using Cortex AI Functions

Try with Cortex Code (before running cells):

Explain what the AI_COMPLETE function does in this notebook

After running all cells:

Summarize the sentiment analysis results shown in the notebook output

Step 4: Build ML Pipeline with Cortex Code

This is where Cortex Code shines - build an entire ML pipeline through conversation.

Note: To use Snowflake Intelligence in Step 6, you must complete either Option A or Option B to register an ML model.

Option A: Run pre-built code

1. Open the TRAIN_ML_MODELS notebook and run all cells

Option B: Build it yourself with Cortex Code

1. Create a new blank notebook with these settings:
   • Notebook location: FSI_DEMO_DB / ANALYTICS
   • Runtime: Run on warehouse
   • Query warehouse: FSI_DEMO_WH
   • Notebook warehouse: FSI_DEMO_WH
2. Install required packages using the Packages selector (top of the page):
   • lightgbm
   • scikit-learn
   • matplotlib
   • seaborn
   • statsmodels
   • snowflake-ml-python
3. Add a Python cell with the following imports and run it:

# Import python packages
import pandas as pd

# We can also use Snowpark for our analyses!
from snowflake.snowpark.context import get_active_session
session = get_active_session()

4. Open Cortex Code (bottom-right icon)
5. Use these prompts one at a time, running the generated code after each:

Note: You can start with Prompt 1 immediately - Cortex Code will explore the FSI_DATA table directly.

Prompt 1: Feature Engineering

Using FSI_DEMO_DB.ANALYTICS.FSI_DATA table, help me construct features with returns: the last 1 day return using close price, return from t-4 to t-1, return from t-9 to t-5, return from t-20 to t-11, and return from t-62 to t-21. Also construct the predictive variable as future return from t+2 to t+6. Take the log across all return variables. Keep as panel data where ticker is a column.

After Cortex Code generates the code:

1. Copy the SQL into a new SQL cell
2. Click on the cell name (e.g., cell2) in the top-left corner and rename it to features_df
3. Run the SQL cell

Prompt 2: Train ML Model

Using features_df (a SQL cell result - use .to_pandas() to convert), train a predictive LightGBM model with L2 metric. Do walk-forward training on a quarterly basis. For each test quarter: Train on all quarters < (Q-2), Validate on (Q-2, Q-1), Test on Q. Enforce strict cutoffs so rows needing returns beyond the split end are dropped (no look-ahead).

Prompt 3: Backtesting

Test if the strategy works starting 2021. For each portfolio construction, generate forecasts on Tuesdays. At Wednesday close, go long top-5 and short bottom-5 by predicted return (equal weight). Hold through Thu to next Wed (the t+2..t+6 window). Transaction cost: 3.0 bps one-way via weekly turnover. Show Information Ratio (before/after costs), Max drawdown, and plot the equity curve.

Prompt 4: Register Model in Snowflake

Register the final model in Snowflake Registry with model name "FIS_STOCK_RETURN_PREDICTOR_GBM", sample input of 100 rows, target_platforms=["WAREHOUSE"], and method_options for predict with case_sensitive=True.

Prompt 5: Verify Model Registration

Show me all models registered in FSI_DEMO_DB.ANALYTICS schema

Explore the code further:

Explain how the walk-forward validation prevents look-ahead bias

Compare model performance across different feature combinations to identify which return windows matter most

Why might the model be overfitting? Suggest fixes

Step 5: Run Notebook - Create Cortex Components

1. In Snowsight, navigate to Projects > Notebooks
2. Open the CREATE_CORTEX_COMPONENTS notebook
3. Run all cells to create Cortex Search, Semantic View, and the Agent

Step 6: Test the Agent

Navigate to AI & ML → Snowflake Intelligence and try these example questions:

ML Predictions (StockPerformancePredictor):

Give me top 3 vs bottom 3 trade predictions for the next period

What are the model's top stock picks right now?

Show me the bottom 5 predicted performers

Structured Data Queries (Cortex Analyst):

Which companies have the highest sentiment score?

What is the average sentiment score by company?

Show me sentiment trends over time for MSFT

Semantic Search (Cortex Search):

Search for companies with concerns about margins

Find earnings calls where analysts discussed supply chain issues

Search for bullish commentary about revenue growth

Combined Analysis:

Compare the top predicted stocks with their analyst sentiment scores

Let's observe if any high sentiment in the bottom 3 performers, and summarize the qualitative insights from the earnings call that shows top sentiment

Show me the top 5 predictions and search for any negative sentiment in their earnings calls

Email Reports (SendEmail):

Note: Email functionality requires your Snowflake user to have a verified email address. Verify your email in Snowsight: User menu → Setting → Profile → Verfify Email.

Send me an email summary of today's top stock picks

Email me a report of companies with sentiment scores above 8

Send an email with the top 3 vs bottom 3 predictions

Cortex Code Power Moves

Beyond following the guided steps, here's what Cortex Code can do:

Explore & Understand

What tables exist in FSI_DEMO_DB.ANALYTICS? Describe each one.

Explain this notebook cell by cell

What's the schema of the AI_TRANSCRIPTS_ANALYSTS_SENTIMENTS table?

Debug & Fix

This cell is throwing an error - help me fix it

Why is my model prediction returning NULL?

The query is slow - how can I optimize it?

Analyze Results

Summarize the model's feature importance

Compare actual vs predicted returns for Q4

Which stocks had the biggest prediction errors?

Cleanup

To remove all demo objects, run the teardown script:

1. In Snowsight, navigate to Projects > Workspaces
2. Create a new SQL file and copy the contents from scripts/teardown.sql
3. Run the script

What's Next?

You've built an end-to-end AI-powered quantitative research pipeline entirely within Snowflake. From here, you can:

• Expand coverage - Add more companies beyond the DOW 30
• Add new features - Use Cortex Code to add technical indicators (RSI, MACD, Bollinger Bands)
• Improve the model - Experiment with different ML algorithms or hyperparameters
• Build dashboards - Create a Streamlit app to visualize sentiment trends
• Automate updates - Schedule daily predictions with Snowflake Tasks

The best part? You can use Cortex Code to help with all of it—just describe what you want to build.

Resources

• Cortex Code Documentation
• Cortex AI Functions
• Cortex Search
• Cortex Analyst
• Snowflake Intelligence
• Snowflake ML Model Registry
• Snowflake Notebooks

License

Copyright (c) Snowflake Inc. All rights reserved.

The code in this repository is licensed under the Apache 2.0 License.