Quarter-Level Profit Prediction for Delta Airline Using Macroeconomic Indicators

Authors: Zilin Pan, Billy Zhang
Study Period: 2003 – 2025

---

1. Introduction

Overview

This study analyzes a comprehensive dataset spanning from 2003 to 2025, integrating internal airline operational metrics—such as Total_Passengers, Total_Fuel_Cost, and Delay—with external macroeconomic indicators including GDP_Growth and Oil_Price. The data was standardized to a quarterly frequency using a PeriodIndex to ensure temporal consistency for time-series modeling.

Predictive Task

The primary objective is to forecast Total_Profit. By employing multivariate time-series modeling, this research investigates how exogenous variables influence airline profitability and captures the seasonality and inertia of profit fluctuations through historical look-back windows.

Summary Findings

• Strongest Predictors: Oil Prices and GDP Growth are the most significant drivers of profitability.
• Model Characteristics: The optimized model exhibits a "short-term memory" characteristic, with an optimal window_length of 2 quarters, suggesting that aviation profits are highly sensitive to immediate environmental shifts.

---

2. Data Description & Processing

2.1 Internal Operational Metrics

Data was aggregated from multiple sources, including BTS (Bureau of Transportation Statistics) web scraping:

• Passenger Volume: Aggregated from monthly data into quarterly Total_Passengers.
• Fuel Costs: Extracted from BTS fuel tables using pd.read_html and aggregated into Total_Fuel_Cost.
• Operational Delays: Statistical count of quarterly delay incidents.
• Load Factor: Measuring the efficiency of seat utilization across the fleet.

2.2 Macroeconomic Indicators

External data was fetched via the FRED (Federal Reserve Economic Data) API:

• GDP Growth: Reflecting the overall health of the economy and travel demand.
• Oil Price: A primary driver of operational expenditure and fuel cost volatility.

---

3. Modeling Approach

3.1 Feature Engineering

The model incorporates historical dependencies using a specialized technical pipeline:

• Standardization: StandardScaler to ensure all features are on a comparable scale.
• Polynomial Expansion: PolynomialFeatures to capture complex, non-linear interactions between economic indicators and operational costs.

3.2 Machine Learning Framework

As implemented in the code.ipynb, the study utilizes:

• Primary Model: RandomForestRegressor.
• Optimization: GridSearchCV for hyperparameter tuning (optimizing n_estimators, max_depth, etc.).
• Architecture: Scikit-Learn Pipeline to automate data transformation and prevent data leakage during training.

---

4. Results and Analysis

The research demonstrates that external economic conditions provide substantial predictive power beyond historical profit data alone:

1. Fuel Sensitivity: Oil price fluctuations have a nearly immediate and significant impact on quarterly profit margins.
2. Economic Correlation: GDP fluctuations serve as a robust leading indicator for passenger demand and premium travel volume.
3. Model Performance: The Random Forest approach successfully captured the cyclicality and volatility inherent in the aviation sector.

---

5. Conclusion and Next Steps

Conclusion

This framework confirms that tree-based ensemble methods are highly suitable for capturing the non-linear relationships in industries characterized by cyclical demand and cost volatility.

Future Improvements

• Lagged Variables: Incorporating lagged oil-price variables to better capture delayed effects arising from fuel hedging contracts.
• Efficiency Metrics: Introducing "Profit per Available Seat Mile" (CASM/RASM) to control for airline scale expansion.
• Scenario Analysis: Developing stress-testing tools for extreme market conditions, such as sudden oil price spikes or recessionary demand collapses.