Airline Industry — Projected Recovery
Using deep learning and DCF to predict Air Canada’s stock prices three years into the future
The airline industry has been in a slump the world over since the commencement of lock-downs across the globe and the spread of the unique coronavirus.
Lately, there has also been much news about Warren Buffet selling his stakes in various airlines. Contrary to his actions, value investors are encouraging many to hold onto airline stocks, especially those with a long investment horizon and an appetite for risk.
The barrage of news piqued my interest and I decided to run some of my own tests to validate the news which I was seeing and hearing.
In order to better understand (on a personal level) and intertwine the airline industry with the performance of Air Canada, I looked at aircraft movements in Canada, used an LSTM model to predict the future stock prices of Air Canada and eventually used a Discounted Cash Flow (DCF) method to arrive at the intrinsic share value.
Forecasting Aircraft Movements in Canada
I collected data from Stat Canada about aircraft movement from January 2015 (major airports) up to April 2020 and compiled it. Subsequently, I created an index linked to the aircraft movements in July 2018 as it was the highest number — this would allow me to gauge the level of aircraft movements in the future with those in a bullish market.
First, I created a monthly average for all the twelve months in a year and used that to create a seasonality index for every month (i.e. Monthly Average of a Month / Average of all Months). I then calculated the average aircraft moments for every month and (index values) and check the difference of those values with those of ‘covid’ the affected months of Jan — May 2020. The index value of April 2020 was 10 which correlates with reports of airline cutting between 85–90% of their capacity for the covid affected months.
Further, I assumed that the effects of covid and the international travel restrictions and lock-downs will commence to open up — at least for businesses and other important drivers of economy by end June 2020. I graduated the index difference downward by 2.5 points to cater to conservative growth estimates considering that aircraft will have to undergo maintenance checks and pilots will have to regain their ‘currency’ before being pressed into service.
My intention was to forecast the values up to the completion of a stabilized cycle of business as this would indicate the normalization of aircraft movements.
I found that aircraft movements reach ‘pre-covid’ levels in May 2022 and stabilize by the end of December 2023.
Predicting Future Stock Prices via Python
I used a simple uni-variate LSTM model in Python to predict the future prices of Air Canada (up to December 2023). Find the notebook here (you can upload your own datasets to run the tests).
I trained and validated the model on the data set of the airline stocks during the effects of September 9/11 faced by the airline industry. I chose this data set for two reasons. Firstly, it was similar to the present situation where customer/public is afraid to fly. Secondly, the airline industry was hit the heaviest during these years in an almost force majeure which led to a change in the way the world flies, with the introduction of various safety regulations and checks. As this would allow for the necessary delay in pickup of business, I felt that this would help predict a better fit for the post covid economy which is likely to face similar limitations.
The results indicate that the stock price of Air Canada will hover at $16 in December 2023, which is after stabilization and recovery of the economy.
Additionally, in order to gauge the present prices of the stocks of Air Canada, I used a basic DCF method (WACC — 5.7%, EV/EBITDA — 6x) and graham’s method (TTM EPS, adjusted to reflect present AAA bond yields, with a 4% growth rate) to calculate the intrinsic value of the Air Canada stocks.
The intrinsic value stood at $9.90 and $12.3 for the respective calculation. The Python model predicts a bottom of $3.5 before it shows upward movement again.
Results and Interpretation
Economy — Assuming that aircraft movements represent the health of the aviation industry (which in turn represents the health of the economy), as most travel is undertaken via discretionary finances, it seems that the economy will stabilize its growth during Q3 of 2021 and will touch pre-covid levels by Q2/Q3 of 2022 before completely stabilizing in Q1 2024. Interestingly, IATA — an association of global airlines has predicted that passenger demand will recover only by 2024, somewhat similar to my model with a December 2023 forecast.
Airline Industry — While the modelling has been done in a fairly rudimentary manner, it seems that the airline industry will take the best part of three years to recover. Additionally, it will be interesting to see the change in behaviour of consumers and how it will affect air travel, as this will largely affect the future of the travel and hospitality industry. Also interesting to see will be how airline companies scale their operations to match the higher demand after the present staff cuts and storage of aircraft. Various checks and re-currency training for aircraft and aircrew after a long layoff (usually six months) will require that airline companies will have to constantly feel the pulse of the economy so that they can plan aircrew training and aircraft maintenance checks to capitalize on the crest of the cycle in a few years time.
Air Canada Stock — The stocks of Air Canada are overvalued by 35% as per the last closing price ($14.62 on May 15 2020) when compared to the average of the intrinsic value arrived via DCF and Graham method. This perhaps explains why Warren Buffet, noted for finding value stocks dumped his airline holdings earlier this year. Thus, unless you have an investment horizon of 7–10 years and are young/or have a high appetite for risk, the airline stocks (Air Canada at least) seem to be a losing proposition.
I want to clarify that the methods and assumptions used for prediction do not account for various nuances. All views/opinions are my own and the work undertaken has been only to help myself have a slightly better appreciation of the impact of the pandemic on the economy. Additionally, the predictions via Machine Learning and the DCF valuation must not be viewed in isolation.
