Nixtla Event Impact Modeler

Nixtla Event Impact Modeler

Quantifies the causal impact of exogenous events on contract prices using TimeGPT forecasting and CausalImpact analysis.

Overview

This skill analyzes how external events (promotions, natural disasters, policy changes) affect contract prices over time. It combines historical price data with event details to quantify causal impacts using MCMC-based counterfactual modeling and TimeGPT forecasting. The skill produces impact estimates, adjusted forecasts, and visualizations for event-driven price changes.

Use cases: Promotion effectiveness analysis, disaster impact quantification, policy change assessment, pricing anomaly investigation, event-aware forecasting.

Prerequisites

Environment:

• NIXTLA_TIMEGPT_API_KEY (required for TimeGPT forecasting)

Dependencies:

pip install nixtla pandas causalimpact matplotlib

Input requirements:

• prices.csv: Contract prices with columns ds (datetime), price (numeric)
• events.csv: Event data with columns ds (datetime), event (string description)

Instructions

Step 1: Prepare data

Load and validate contract price and event data using the data preparation script.

python {baseDir}/scripts/prepare_data.py \
  --prices prices.csv \
  --events events.csv \
  --output-prices prepared_prices.csv \
  --output-events prepared_events.csv

To create sample data for testing:

python {baseDir}/scripts/prepare_data.py --create-sample

Script actions:

• Loads CSV files with datetime parsing
• Validates required columns (ds, price/event)
• Renames columns to Nixtla standard (y for price)
• Adds default unique_id if missing
• Outputs prepared CSVs for analysis

Step 2: Configure model

Define event windows and mark treatment/control periods in the price data.

python {baseDir}/scripts/configure_model.py \
  --prices prepared_prices.csv \
  --events prepared_events.csv \
  --window-days 3 \
  --output configured_prices.csv

Script actions:

• Defines event periods with configurable window (default: 3 days before/after)
• Validates event dates fall within price data range
• Creates treatment column (1=treatment period, 0=control period)
• Outputs configured DataFrame with treatment markers

Parameters:

• --window-days: Event window size in days (default: 3)

Step 3: Execute analysis

Run CausalImpact analysis with TimeGPT forecasting to quantify event effects.

python {baseDir}/scripts/analyze_impact.py \
  --prices configured_prices.csv \
  --events prepared_events.csv \
  --niter 1000 \
  --window-days 3 \
  --output-impact impact_results.csv \
  --output-forecast adjusted_forecast.csv \
  --output-summary causal_summary.txt

Script actions:

• Defines pre-intervention and post-intervention periods
• Runs CausalImpact MCMC analysis (configurable iterations)
• Calculates absolute and relative event effects
• Generates TimeGPT adjusted forecasts
• Outputs impact metrics, forecasts, and summary report

Parameters:

• --niter: MCMC iterations for CausalImpact (default: 1000)
• --window-days: Event window size (must match Step 2)

Step 4: Generate report

Create visualization and markdown report summarizing the analysis.

python {baseDir}/scripts/generate_report.py \
  --impact-results impact_results.csv \
  --adjusted-forecast adjusted_forecast.csv \
  --causal-summary causal_summary.txt \
  --output-plot impact_plot.png \
  --output-report impact_report.md \
  --title "Event Impact on Contract Prices"

Script actions:

• Generates time series plot with actual prices, forecasts, and treatment periods
• Creates markdown report with impact metrics, CausalImpact summary, and methodology
• Outputs high-resolution PNG and structured markdown report

Output

Generated files:

• impact_results.csv: Event impact metrics (absolute effect, relative effect, average price)
• adjusted_forecast.csv: TimeGPT forecasts with actual prices and predictions
• causal_summary.txt: CausalImpact statistical summary
• impact_plot.png: Time series visualization with treatment periods highlighted
• impact_report.md: Comprehensive markdown report with all results

Impact metrics:

• Absolute effect: Total price change attributable to events
• Relative effect: Percentage change relative to mean price
• Counterfactual forecast: What prices would have been without events

Error Handling

Examples

Example 1: Promotion impact analysis

Scenario: Quantify price increase during promotional campaign.

Input:

• prices.csv: Daily prices for 30 days
• events.csv: Single promotion event on day 15

Command sequence:

python scripts/prepare_data.py --prices prices.csv --events events.csv
python scripts/configure_model.py --prices prepared_prices.csv --events prepared_events.csv --window-days 5
python scripts/analyze_impact.py --prices configured_prices.csv --events prepared_events.csv --niter 2000
python scripts/generate_report.py --impact-results impact_results.csv --adjusted-forecast adjusted_forecast.csv

Output: impact_results.csv shows 15% relative price increase during promotion period.

Example 2: Natural disaster impact

Scenario: Assess price drop following natural disaster.

Input:

• prices.csv: Weekly prices for 52 weeks
• events.csv: Disaster event on week 26

Command sequence:

python scripts/prepare_data.py --prices prices.csv --events events.csv
python scripts/configure_model.py --prices prepared_prices.csv --events prepared_events.csv --window-days 7
python scripts/analyze_impact.py --prices configured_prices.csv --events prepared_events.csv
python scripts/generate_report.py --impact-results impact_results.csv --adjusted-forecast adjusted_forecast.csv --title "Disaster Impact Analysis"

Output: impact_report.md documents price recovery timeline and total economic impact.

Resources

Scripts (all in {baseDir}/scripts/):

• prepare_data.py: Data loading and validation with argparse CLI
• configure_model.py: Event period configuration and treatment/control marking
• analyze_impact.py: CausalImpact + TimeGPT analysis engine
• generate_report.py: Visualization and markdown report generation

Documentation:

• Nixtla TimeGPT API
• CausalImpact Python
• StatsForecast Documentation