CHAPTER V

DISCUSSION

Even though the results did not fully support the hypothesis, there are several different scenarios that affect each of the result sets.  By far the most accurate and most basic set of predictions was from the ATL - MCO route.  With a mean absolute deviation of 9.37%, this is the only result set that supported the hypothesis within the boundaries set; however, by categorizing the results by time frame, an interesting trend emerges which will continue through all five tested markets.  In the six months before September 2001, the ATL - MCO market predictions proved incredibly accurate with only 4.91% error.  September 11, 2001, triggered an 18.92% prediction error throughout the next four quarters, a high value which was easily compared to the other leisure market of JFK - MCO with 31.18% prediction error.  It is apparent from these two percentages that September 11, 2001, had a relatively larger effect on leisure markets than it did on the traditional business markets of DFW - ORD and ATL - LAX, with errors of 16.82% and 11.26% respectively.  The LGA - DCA market is almost exclusively used by business travelers; however, with DCA airport being closed after September 11, 2001, and the restrictions which followed months after the reopening, the prediction error of 47.44% during this period is undoubtedly affected heavily by this.

   The success of the predictions on the ATL - MCO route did not seem to be as attributable to a change in independent variables; rather, the passenger boardings on that route seem to be more closely aligned to the trending boardings due solely to the previously recorded values.  Because of this, a simple linear regression forecasting technique should produce similar results.  The highest r-squared values (coefficient of determinations) were seen in daily seats, average frequency, and average aircraft size, which suggests that capacity is the determining factor to demand between Atlanta and Orlando.  While the fares did affect passenger boardings (r-squared value of .16 for all fares and .15 for nonstop fares), they were used as more of a reaction to the capacity on this route to maintain constant load factors and market share.  This illustrates the lack of pricing power airlines are having on routes such as ATL - MCO in which capacity controls price, rather than price controlling capacity.

The other market operating out of Atlanta, ATL - LAX, saw the introduction of low cost carrier JetBlue Airways in Q2, 2003.  Prior to this time, the route was predominately business travelers, as was apparent by the muted response to September 11.  With no previous introduction to the ATL - LGB market, the neural network did not accurately predict the effects of JetBlue's; however, only quarter later it did compensate.  The large dip in predicted passenger boardings during Q1, 2004 came as a result of JetBlue leaving the market, but fares did not react accordingly due to the additional capacity Delta Airlines had added to the market.  This illustrates a shift in route dynamics from an inelastic demand scenario before JetBlue's entry to an elastic demand scenario after its entry.  Once again price is being dictated by capacity, and with Airtran's small, yet persuasive, presence in this market, this situation will see no quick resolution.

   Also of interest to the ATL - LAX market are the strong correlations between passengers and daily seats (.86) and average frequency (.88) while a low correlation between passengers and aircraft size (.06).  This is most likely due to the distance of the route on which smaller aircraft tend to not have the range to complete such a flight.  Thus, in order to increase capacity, the airline is forced to add additional flights rather than increase aircraft size because the larger aircraft are already on this route.  The extra aircraft will yield no large change in average aircraft size, yet both the frequency and the daily seats will change significantly.  The same is not true, however, on shorter flights such as LGA - DCA, where average aircraft size had a much higher correlation of .23.

   Another route which showed leisure market characteristics was JFK - MCO.  Even though the results did not predict prior to the year 2000, the effect of JetBlue's entrance into this market in Q2, 2000 is clearly evident.  Once again the neural network seems to take one full year to stabilize predictions after a major event, and can be seen three separate times on this market:  after JetBlue's entry in Q2, 2000, after September 11 in Q3, 2001, and after Song's introduction of service in Q3, 2003.  Considering that all three events occurred within the scope of these predictions, an overall error of 17.44% does not seem extraordinary, even though it is the highest of the set.  Once again there are high correlations to capacity variables such as daily seats (.96) and average frequency (.92).  Interestingly enough, the passenger boardings between JFK and MCO were most affected by the fares between LGA and MCO than the fares on any of the competing markets.  LGA - MCO non-stop fares showed a correlation of .52 while JFK - MCO non-stop fares had an r-squared value of only .08.