50 years on, should we still be studying CAPM?
A company’s share price could reasonably be summarised as the discounted value of its future cash flows in the form of earnings. If an analyst gets this prediction right, he will be able to accurately assess the merits of its current price. A key question therefore, is “what models and indicators will the analyst use to help him predict these future returns?”
The Capital Asset Pricing Model (CAPM) has been a primary risk / return indicator for securities since its introduction in 1964 by John Lintner, William Sharpe, Jack Treynor and Jan Mossin. The model developed from the earlier work of Harry Markowitz relating to modern portfolio theory.
In addition to the expected return of a risk free asset, the CAPM formula takes into account the relationship of a security to the non-diversifiable risk of the market i.e. its beta (β) and calculates the specific premium this should provide. The total return of a security according to CAPM is thus expressed in the following formula:
Eri = Rf + β(Erm – Rf)
Where:
Eri = Expected return of security
Rf = Risk free rate of return
β = Beta of security relative to the market
Erm = Expected market return.
CAPM has been accepted academically since its introduction some 50 years ago and will dominate any conversation in an investment textbook as a key indicator of returns relative to risk. However, empirical evidence suggests that the model has its issues. A lot of these arise from some key assumptions applicable to the base theory:
- All investors are risk averse by nature
- Investors have the same time period to evaluate information
- There is unlimited capital to borrow at the risk-free rate of return
- Investments can be divided into unlimited pieces and sizes
- There are no taxes, inflation or transactions costs
At first sight, these assumptions appear unrealistic and over-simplified relative to market conditions. Back testing of the model has also highlighted some concerns too.
Research from the late 1970s found that CAPM had shortcomings in terms of shares with high earnings yields in that they tended to perform more positively than predicted by the model. More research in the early 1980s (Banz et al) found that small stocks as measured by market capitalization outperformed what CAPM would have predicted.
Fama and French latterly attempted to address these issues by the introduction of a three-factor model using risk premiums for small cap and value stocks as well as the market risk measured by Beta. Back testing found that the accuracy of predicting returns went from around 60% to approximately 90%. Since then, further models have been created to refine or even replace the basic model.
Thus although studies continue, evidence so far suggests that many of the fundamental indicators followed by analysts contain information not captured within the beta of a stock and therefore not predicted using CAPM.
With so much research attacking the validity of CAPM, why would it still be so widely recognized, studied and accepted?
It may be the validity of the research itself. Has this truly captured the behavior of securities within the marketplace or has it been so narrowly defined that the sample stocks used have provided indicators that are not fully representative of all stocks?
Sorting firms on metrics like price / book or price / earnings ratios exposes investors’ subjective reactions, which tend to be highly positive in good times and overly negative in bad. Investor behaviour often results in an over-forecasting of past performance too. This can mean stock prices being too high for high price / earnings firms (growth stocks) and too low for low P / E firms (value stocks). Once the cycle is complete, the results often mean higher returns for value stocks and lower returns for growth stocks.
50 years on from its introduction, CAPM is still one of the most widely studied and accepted pricing models but in that timespan, it has generated several critics. Its assumptions have been derided from the outset as unrealistic in real market conditions. Further studies have demonstrated that factors such as market size and company earnings levels provide an argument to add to the basic model’s formula or even ignore it altogether.
However, perhaps one reason why the model maintains its notoriety is its pure simplicity and the premise that it can work extremely well as an overall indicator of predicted returns if the stock has a high association of returns relative to the market return indicator being used. Only time will tell how long its validity will last.