Analysis of the Modern Financial Regression Model: CAPM and
Fama-French
Yucheng Lin
College of Arts and Sciences, University of North Carolina at Chapel Hill, Chapel Hill, U.S.A.
Keywords: CAPM, FF3, FF5, Financial Pricing, Expected Returns.
Abstract: Contemporarily, financing pricing models are widely adopted in assets evaluations. This study will discuss
the development and empirical testing of the CAPM, Fama-French Three-Factor Model (FF3), and the Five-
Factor Model (FF5) in the context of modern asset pricing. The FF3 model improved upon the traditional
CAPM by introducing size and value factors to better capture variations in stock returns. However, its inability
to account for differences in profitability and investment behaviours led to the creation of the FF5 model,
which adds profitability and investment factors. Empirical evidence suggests that while the FF5 model
generally outperforms the FF3 model in explaining stock returns, particularly in the U.S. market, its
performance is less consistent in other markets, such as China and Japan, indicating its limitations in diverse
economic and regulatory environments. The review highlights the FF5 model's potential factor redundancy,
inconsistent results across markets, and limited ability to capture certain stock behaviours as areas requiring
further development. Future research directions include integrating artificial intelligence and behavioural
factors to enhance the model's predictive power and applicability. This research concludes that while the FF5
model represents a significant advancement in asset pricing.
1 INTRODUCTION
In broad terms, finance revolves around the study of
asset risk and return through mathematically lens. In
the early 1960s, American economist William Sharpe
and his colleagues revolutionized modern finance
through the Capital Asset Pricing Model, providing
foundation and framework for valuation of assets.
The pioneering financial model applied widely in the
20th century for and William Sharpe was awarded the
Nobel Price of Economic Science in 1990. In addition
to the Markowitz Model assumptions on risk aversion
of investor, the one-factor model added the
assumptions of complete agreement of investors and
risk-free rate borrow and lend (Fama & French, 2004).
CAPM asserted a linear relationship between asset
returns and market risk (Ross, 1978). Despite its
theoretical simplicity and strong explanatory power,
CAPM has faced criticism in practical applications,
particularly when dealing with real-world markets
characterized by incomplete information and
heterogeneous investors (Fama, 1970; Epps, 1976).
To address the limitations of CAPM, Fama and
French proposed FF3 model, introducing two
additional factors to explain anomalies in stock
returns that CAPM could not account for. The FF3
model demonstrated significant improvements in
various empirical studies, particularly in explaining
the excess returns of small-cap and value stocks
(Fama & French, 1993).
Fama and French later introduced the FF5 which
improves upon FF3, adding two additional factors to
the original three. FF5 aims to reflect the drivers of
stock returns, providing a more robust asset pricing
tool for both academic and practical use (Fama &
French, 2015). Empirical tests on developed global
markets suggest an improvement to FF3 and that the
regional models outperform global model (Cakici,
2015). This study will systematically review and
analyse the existing methodologies on the CAPM,
FF3, and FF5 models to explore their theoretical
foundations, applications, and performance across
different market environments.
This paper will first examine the history and
theoretical expansion of the modern financial models
and then assess their performance in empirical studies.
Then the paper will compare the application
outcomes of FF3 with FF5 across different markets.
Through this literature review, I hope to highlight the
limitations and prospects of these models in modern
314
Lin, Y.
Analysis of the Modern Financial Regression Model: CAPM and Fama-French.
DOI: 10.5220/0013224200004568
In Proceedings of the 1st International Conference on E-commerce and Artificial Intelligence (ECAI 2024), pages 314-318
ISBN: 978-989-758-726-9
Copyright © 2025 by Paper published under CC license (CC BY-NC-ND 4.0)
financial markets, incorporating the models with new
emerging technologies of Artificial intelligence.
2 DEVELOPMENT FROM CAPM
TO FAMA-FRENCH MODEL
2.1 History of CAPM
In modern finance, CAPM remains as a simplistic yet
fundamental model for valuation of asset and
estimation of portfolio performance. CAPM,
introduced by Sharpe and Lintner, builds upon
Markowitz's mean-variance optimization theory
developed in the 1950s. Markowitz’s model assume
investors to minimize the variance and maximize the
return rate of a portfolio. The model indicates prices
of all assets reflect their market risk, represented by
their beta coefficients. It distinguishes between
systematic risk and unsystematic risk (risk specific to
an individual asset), assuming that investors
concerned with systematic risk primarily. It assumes
every investor's portfolio follows the mean-variance
optimization strategy and can borrow or lend
unlimited amounts at the risk-free rate (Ross, 1978).
Based on the assumptions, CAPM concludes on a
linear relationship between the expected return and
beta:
𝐸
𝑅
= 𝑅
+ 𝛽
(𝐸
(
𝑅
𝑅
) (1)
Beta represents the systematic risk of the market and
depends on historical returns and market returns,
𝛽
=
(
,
)
(
)
(2)
Beta also represents the slope of linear regression
between the expected return and market return which
provides information on sensitivity of individual
stock given the fluctuating market return. Under the
assumption of CAPM, the beta of riskless asset equals
zero and the expected return of the portfolio must
equal to the risk-free rate.
While CAPM provides a simplistic algebraic
method to valuate asset, it has limitations due to
assumptions on investors behaviors. It assumes an
efficient market where all investors behave rationally,
and they can borrow or lend at the risk-free rate.
Contrary to this idealized assumption, market
information is not fully transparent, transaction costs
exist, and borrowing is not always possible at the risk-
free rate, thereby weakening CAPM's effectiveness in
real-world applications (Ross, 1978; Bartholdy &
Peare, 2005). The assumption of investors caring only
about mean-variance also reflects the flaw of CAPM.
It is likely for investors to care for income in addition
to the mean-variance analysis, thus undermining the
portfolio’s return variance as it misses such variables
(Fama & French 2004).
2.2 CAPM TO FF3
Academia recognized the limitations to a single-
factor model in valuating asset. Eugene Fama and
Kenneth French introduced the FF3 in 1993, which
added two additional factors—firm sizeand book-to-
market ratio—to capture the effects of size and value
on stock returns:
𝐸
(
𝑅
)
= 𝑅
+ 𝛽
𝐸
(
𝑅
)
𝑅
+ 𝛽
(
𝑆𝑀𝐵
)
+
𝛽
(
𝐻𝑀𝐿
)
(3)
Intuitively, higher expected returns for small-cap
stocks represented by the size factor will compensate
investors for the additional risks associated with
smaller firms, which may include higher volatility,
less liquidity, and higher default risks. Small firms
often have less access to capital, fewer resources to
weather economic downturns, and are generally
considered riskier investments compared to their
larger counterparts. Investors demand a risk premium
for holding that stocks, leading to higher expected
returns relative to large-cap stocks (Fama & French
1992). FF3’s logic builds on the premise that stock
returns are influenced not only by their sensitivity to
overall market risk but also by their exposure to these
additional risk factors. By including these two
additional factors, FF3 provides a more nuanced
explanation of stock returns, reflecting the risk
premiums investors demand for holding small-cap
and value stocks due to their perceived higher risk.
2.3 Development of FF5
In 2014, Eugene Fama and Kenneth French extends
the original model again with additional factors of
profitability and investment. FF5 extends on logic of
FF3 using factors of RMW and CMA,
𝐸
(
𝑅
)
= 𝑅
+ 𝛽
𝐸
(
𝑅
)
𝑅
+ 𝛽
(
𝑆𝑀𝐵
)
+
𝛽
(
𝐻𝑀𝐿
)
+ 𝛽
(
𝑅𝑀𝑊
)
+ 𝛽
(
𝐶𝑀𝐴
)
+ 𝛽
(4)
Here, RMW calculates the difference in returns
between a portfolio of stocks with robust and weak
portfolio. Profitability is typically measured by
operating profitability. Research shows that firms
Analysis of the Modern Financial Regression Model: CAPM and Fama-French
315
with higher profitability tend to have higher stock
returns, as they are often perceived as more stable and
less risky investments. Profitable companies generate
stronger cash flows and are better able to weather
economic downturns, which justifies a risk premium
on their returns (Fama & French, 2015). CMA
calculates the difference in returns between firms that
follow low asset growth and those that adopt high
asset growth. The CMA factor captures the
observation that firms with lower levels of asset
growth tend to have higher future returns comparing
to those with higher levels of asset growth. This can
be attributed to the tendency of firms that invest
aggressively to undertake riskier or less profitable
projects, which may lead to lower future profitability
and hence lower expected returns. Conversely, firms
with conservative investment policies are generally
viewed as more disciplined and less risky, resulting
in relatively higher expected returns. If the Betas
explain all cross-sectional variation of the portfolio,
then 𝛽
, or the intercept, equals zero (Fama & French,
2015).
FF5 bases in Dividend Discount Model (DDM)
which suggests that expected returns are linked to
profitability and investment patterns. DDM states
that given the expected dividend 𝐸(𝑑

), the stock
price:
𝑚
=
𝐸(𝑑

)/(1 + 𝑟)

(5)
The relation given by DDM has implications on
the expected return of an individual stock and CMA
and RMW aim to explain cross-sectional variation
under this model. Despite its broader scope, the FF5
model has faced several criticisms. One key criticism
is the potential redundancy of the value factor (HML).
Fama and French found that the HML factor's
significance diminished with the inclusion of FF5
factors, raising questions about its necessity in the
model (Cakici, 2015). Furthermore, the FF5 model's
performance varies across different markets and
portfolio constructions, suggesting that its
applicability might be context dependent (Fama &
French, 2015). The limitations suggest a need for
empirical examination of the FF3 and FF5 Model
performance.
3 APPLICATION AND
DEVELOPMENT OF FAMA-
FRENCH MODEL
3.1 Empirical Stuies
In the study on performance of FF3 and FF5 on
portfolio of ten US sectors, researchers point out that
FF5 explains the variability of sector portfolio better
than FF3 (Sarwar et al., 2017). While both FF3 and
FF5 outperform the benchmark of S&P 500, the result
reveals a higher cumulative return and adjusted R
2
than that of CAPM and FF3. According to the
research, FF5 alpha has higher adjust R
2
than that of
FF3 alpha across all ten US sectors. The return of FF5
portfolio also have 7% higher return and 2% lower
standard deviation compared to the buy-and-hold
S&P 500. With trading on sector ETFs, long-only
FF5 trading strategy has 5.53% of mean return while
buy-and-hold S&P500 only has 2.05%. The
researchers suggest that the factors of RMW and
CMA betas of FF5 likely decrease alpha estimate in
most sectors (Sarwar et al., 2017).
Research on the U.S. market sectors suggests that
the FF5 better explains the variation compared to the
FF3 model, particularly in small-cap, high-
investment portfolios and among highly profitable
companies. However, the FF5 model also shows that
the role of the HML is reduced in certain contexts,
reflecting the ability of the new factors to account for
market anomalies and partially replace traditional
factors.
The valuation model applies significantly
different in global markets and regional markets.
Study shows that RMW and CMA factors in FF5 do
not have significant explanatory power on expected
return in the Chinese A-share stock market (Jiao &
Lilti, 2017). In tracking the performance of FF3 and
FF5 from July 2010 to May 2015, the paper indicates
the distinguishable difference of the Size-B/P
portfolio intercept from zero (Jiao & Lilti, 2017).
Under the assumption where the coefficients (betas)
of the factors explaining the cross-sectional expected
return, the intercept β should equal zero. The non-
zero intercept thus reveals that FF5 Model do not
fully account for the expected return. The researchers
also indicated that coefficient of RMW only have
statistically significant in Size-OP portfolio and that
the coefficient of CMA negatively correlate with
small size-aggressive investment portfolio (Jiao &
Lilti, 2017). Noticeably, the adjusted R
2
of FF5 and
FF3 after running time-series regression on the same
portfolio do not differ dramatically. Probability and
ECAI 2024 - International Conference on E-commerce and Artificial Intelligence
316
investment thus do not have explanatory effect on the
expected return and the two models’ performance do
not have huge difference. The result diverges from the
empirical study of FF3 and FF5 on the US market
where FF5 have higher cumulative return and R
2
.
The difference between the two empirical study
results on FF3 and FF5 performance reveal the
limitation of Fama-French Model in specific regions
or context. Whereas the additional two factors of
CMA and RMW have improved performance in
North America, CMA seems redundant in Europe and
Japan (Fama & French, 2016). The studies highlight
the importance of market-specific empirical testing
and suggest that asset pricing models may require
customization or adjustments when applied to
different economic and regulatory environments.
3.2 Limitations and Prospects
As an empirical model, FF5 has limitations to
measurement of expected return. FF5 model has
shown inconsistent performance across different
markets, highlighting its limitations in capturing local
market dynamics. While the model performs well in
the U.S. market, where it significantly improves upon
the FF3 model, its effectiveness is less pronounced in
other markets, such as the Chinese A-share market
and Japanese markets.
Recent research has begun to explore the
integration of the FF5 model with advanced
technologies and additional factors to improve its
predictive capacity. For example, Mita and Takahashi
(2023) propose a new approach that combines the
FF5 model with artificial intelligence techniques,
such as Gradient Boosting Machine (GBM) and state-
space models, to enhance the accuracy of return
predictions. By using AI-driven predictions in
conjunction with the FF5 factors, this approach can
dynamically adjust to changing market conditions
and better forecast future returns, outperforming
traditional strategies like buy-and-hold or typical
mutual fund approaches for Japanese equities. The
study demonstrates that an AI-enhanced FF5 model
not only retains the strengths of the traditional model
but also provides superior performance, suggesting a
promising future direction for integrating machine
learning techniques into asset pricing models to
handle large datasets and capture complex patterns in
financial markets (Mita & Takahashi, 2024).
As AI and machine learning techniques continue
to advance, there is a significant opportunity to
enhance the FF5 model’s predictive power and
applicability across diverse markets. AI-based
models have the advantage of processing vast
amounts of data more efficiently and identifying
complex, non-linear relationships that traditional
econometric models may miss. By leveraging these
technologies, the FF5 model can potentially evolve
into a more flexible and adaptive tool for predicting
returns in real-time, accounting for rapid changes in
market conditions and investor behavior (Mita &
Takahashi, 2024).
Other researchers have explored the addition of
new factors to the FF5 model to address its current
limitations. Dhaoui and Bensalah expanded the FF5
model by incorporating momentum and investor
sentiment factors, arguing that these additions could
improve the model's ability to capture certain market
behaviors that the standard FF5 model misses. The
inclusion of a momentum factor helps account for the
tendency of stocks to continue moving in their current
direction, while the investor sentiment factor reflects
the impact of psychological biases on asset prices.
Their findings suggest that this enhanced model can
better predict expected returns and explain anomalies
related to small stocks with high investment and low
profitability, i.e., an area where the original FF5
model often falls short. This approach underscores
the potential for further developments that
incorporate behavioral and sentiment-based factors,
offering a more holistic view of asset pricing that
includes both traditional financial variables and
behavioral elements (Dhaoui & Bensalah, 2016).
4 CONCLUSIONS
To sum up, FF3 model marked a significant departure
from the traditional CAPM by additional factors,
which allowed for a more nuanced understanding of
stock returns by accounting for systematic risks
beyond market exposure. FF5 model represents a
further evolution in asset pricing by adding two
additional factors to provide a more comprehensive
model for understanding the variations in stock
returns. Empirical evidence suggests that FF5
generally performs better than FF3 in explaining
returns, particularly in markets like the United States.
The model's performance varies significantly across
different markets. This study recognizes the
explanatory limitation of FF5 as empirical model and
suggest that the emergence of AI or factor that
customized to local financial market will improve the
performance of factors. In conclusion, while FF5
represents a substantial step forward in asset pricing
theory, its mixed empirical results and inherent
limitations indicate that it is not yet the final word on
modelling stock returns. Future developments should
Analysis of the Modern Financial Regression Model: CAPM and Fama-French
317
focus on refining the model by integrating new
techniques and data sources to enhance its robustness,
flexibility, and applicability across global financial
markets.
REFERENCES
Bartholdy, J., Peare, P., 2005. Estimation of expected return:
CAPM vs. Fama and French. International Review of
Financial Analysis. 14(4), 407-427.
Basiewicz, P., Auret, C., 2010. Feasibility of the Fama and
French three factor model in explaining returns on the
JSE. Investment Analysts Journal, 39(71), 13–25.
Dhaoui, A., Bensalah, N., 2017. Asset valuation impact of
investor sentiment: A revised Fama–French five-factor
model. Asset Management. 18, 16–28.
Fama, E. F., French, K. R., 2017. International tests of a
five-factor asset pricing model. Journal of Financial
Economics. 123(3), 441-463.
Fama, E. F., French, K. R., 2015. A five-factor asset pricing
model. Journal of Financial Economics. 116(1), 1-22.
Fama, E. F., French, K. R., 2004. The Capital Asset Pricing
Model: Theory and Evidence. Journal of Economic
Perspectives, 18(3), 25–46.
Fama, E. F., French, K. R., 1992. The Cross-Section of
Expected Stock Returns. The Journal of Finance, 47(2),
427–465.
Jiao, W., Lilti, J. J., 2017. Whether profitability and
investment factors have additional explanatory power
comparing with Fama-French Three-Factor Model:
empirical evidence on Chinese A-share stock market.
China Financ. and Econ. Rev. 5(7).
Mita, D., Takahashi, A., 2024. A New Equity Investment
Strategy with Artificial Intelligence, Multi Factors, and
Technical Indicators. Center for International Research
on the Japanese Economy, 11.
Ross, S. A., 1978. The Current Status of the Capital Asset
Pricing Model (CAPM). The Journal of Finance, 33(3),
885–901.
Sarwar, G., Mateus, C., Todorovic, N., 2018. US sector
rotation with five-factor Fama–French alphas. Asset
Management. 19, 116–132.
ECAI 2024 - International Conference on E-commerce and Artificial Intelligence
318