Kawika Pierson MIT System Dynamics Group 3 nd Year PhD Fall 2008 Albany-MIT PhD Colloquium The Cyclical Nature of Airline Industry Profits
Outline Relevant Literature Reference Mode for Airline Profits Digging Deeper The Model Demand Price Capacity Costs Profit Results
Relevant Literature “Cycles in the sky: understanding and managing business cycles in the airline market” M Liehr, A Groessler, M Klein, PM Milling - System Dynamics Review, 2001 Made for Lufthansa as a guide for strategy Very limited scope (only one feedback loop)
Relevant Literature “System dynamics for market forecasting and structural analysis” James Lyneis - System Dynamics Review, 2000 Commercial jet aircraft industry Focused on use of SD models as forecasts for Jet Orders Proprietary, but potentially similar to our work "Analysis of Profit Cycles in the Airline Industry" 2004 Helen Jiang, R. John Hansman Very simple model, two stocks one feedback loop Control theory perspective
Reference Mode The data for US airline industry profits shows some cyclicality since before deregulation Taken from a presentation by Prof. R. John Hansman and Helen Jiang Nov. 2004
Digging Deeper Profit = Revenue – Costs Revenue = Price * sales in units Costs = unit cost * production Sales is Revenue Passenger Miles Price is the Price of Tickets Production is Available Seat Miles This gives us Profit How does financial reporting effect our modeling?
Capacity – Causal Loop Diagram
Capacity - Model Structure
Third Order Stocks –Cancellation and Mothballing
Forecasting Demand
Correction For Growth
Fitting to Data Get historical data on important stocks Airlines are great for this Airlines.org, MIT Airline Data Project, BTS Set up summary statistics John Sterman’s Book plus MAE, RMSE, %E, Thiel, SSE/M^2 Drive each model sector with historical variables Use Vensim’s model fitting functions Lets walk through this
Summary Statistics
Example of Fitting the Model 1. Open Simulation Control 2. Create a Payoff
Example of Fitting the Model 3. Run “Policy” Negative
Example of Fitting the Model 4. Set Parameters
Example of Fitting the Model 5.
Capacity Fit – Historical Inputs “R^2” MAE/Mean RMSE/Mean Um Uc Us
Demand – Causal Loop Diagram
Demand – Model Structure
Demand Fit – Historical Inputs “R^2” MAE/Mean RMSE/Mean Um Uc Us
Price – Model Structure
Price Fit – Historical Inputs “R^2” MAE/Mean RMSE/Mean Um Uc Us
Costs – Causal Loop Diagram
Costs - Model Structure
Cost Fit – Historical Inputs “R^2” MAE/Mean RMSE/Mean Um Uc Us
Wages – Model Structure
Profits – Model Structure
Wages Fit – Historical Inputs “R^2” MAE/Mean RMSE/Mean Um Uc Us
Real Wages Fit – Historical Inputs “R^2” MAE/Mean RMSE/Mean Um Uc Us
Full model Optimization Move from partial model tests to full model parameterization Fits are slightly worse, parameters more believable MAE/Mean RMSE/Mean
Full model Optimization MAE/Mean RMSE/Mean
Parameters More Believable In Partial Model Test SLAT = 0.05 TAC = 1 Theoretically should be very similar In Full Model Parameterization SLAT = 0.18 TAC = 0.19 Time to Adjust Prices Partial = 0.05 Full =0.64 Sensitivity of Price to Cost Partial = 3 Full = 0
Profits Still Questionable
Conclusions Growth Correction Partial Model Tests with Historical Inputs Cyclical Nature not alleviated by Cancellations or Mothballing Standard SD Structures fit the industry reasonably well More dynamics exist in the real system Comments? Questions?