CRT 2 M2

1. In the following table, the results of 6 models attempt to explain a dependent variable
of interest, yy.
Part I: Provide a thorough, rigorous analysis of which of the models is preferred.

Model 2 is preferred from among the models.

Firstly. lets us look at the coefficients from the regression. The βcoefficients in Model 2 are
not close to Zero as against all the others which have atleast 1 coefficient that is close to
Zero. Model 1 has a lower explanatory power (R Squared of 37.75%) as compared to Model
2 (41.48%).

Secondly, evaluating closely the relevance, or statistical significance, of the independent

variables we can look at the t-Statistic (or P-Value). If t-statistic is high 2+ standard error (or
alternatively if the P-Value is low) then the null hypothesis of the coefficient to be ZERO will
be rejected. Model 3, 4 and 6 have atleast 1 p-value that is very low, and so the null
hypothesis of these coefficients to be zero cannot be rejected. Hence only values of the
coefficients from Model 1, 2 and 5 are significant.

Finally looking at F-statistic that evaluates the joint significance of the coefficients, excluding
the constant α Model 2 has the higher F-statistic among the other models. Model 2 is
parsimonious and has fewer regressors (2) as compared with Model 1 and Model 5 with that
aspect. Hence overall Model 2 is a more preferred regression model.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Number of 2 2 3 3 3 4

Regressors
Coefficients 1 None 1 1 1 2

with Value ~ X1 = 0.0804 X2 = 0.3267 X1 = 0.0812 X4 = 0.0121 X1 = -.0825

0
X3 = 0.2324 X1 = 0.0180

p-Value p-value of p-values X4: p- X4: p- p-values X4: p-

coefficient of the 2β value value of the 3β value
(High Values is high are < are <
make null 0.01 and 0.0001 0.57719 0.55999 0.0001 0.36709
hypothesis other is
more likely lower and for Others are Others are Others are
acceptable) than αis < lower < lower < lower <
0.0001 0.00003 0.0001 0.0003 0.0002
Ho : β = 0
 Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
R2 37.75% 41.48% 37.85% 41.58% 46.38% 46.55%

F-Statistic 42.55 69.81 39.79 46.5 56.51 42.55

Part II: For the preferred model, give an analysis of the likely correlation among the
explanatory variables.

Since Model 2 has more than 1 independent variable, there is a possibility of a

multicollinearity, or correlation between the independent variable. While the data does not
illustrate the correlation constants, or the extent to which changes in the variables cause the
coefficients to change, we can speculate based on some signals from the coefficients and
their parameters. A high multicollinearity will be observed in the following situations:

1. Coefficients with very high standard errors and low significance levels (high p-value for
the coefficient).
2. Coefficients with a wrong ‘sign’ on an implausible magnitude.

Since Model 2 has positive values of the coefficients with low p-values (<0.0001), the extent
of multicollinearity is low.

Notably, Model 6 has 4 parameters, and has one coefficient with a negative value and this
looks implausible when compared to other models like Model 3 which use the same
parameter (X1 = 0.0812).

However, the true extent of the multicollinearity can be measured by ‘Variance Inflation
Factor’ and Condition Number. VIF > 5 shows a potential multicollinearity problem.

2. Use the data file provided: fama_factors_2019.csv

Downloaded CSV. Downloaded separately Adobe Stock prices for the same time
period. Calculated the yearly returns for each time.
FF_ADBE_2019 <- read_excel("D:/Training/MScFE/Econometrics/M2/FF_ADB
E_2019.xlsx",
+ col_types = c("date", "numeric", "numeric",
+ "numeric", "numeric", "date", "numeric",
+ "numeric", "numeric"))
> View(FF_ADBE_2019)

3. Write the Fama-French 3 factor model equation, specifying what each term means.
Fama French model is a 3-factor linear equation, to explain the returns on an asset or a
portfolio. This model extends the single factor CAPM model to include 2 additional factors
for the size differential of the asset
E(r) – Rft = α + β1 *(Km - Rft) + β2 *(SMBt) + β3 *(HMLt)

Rf: Risk free rate term. This is usually the rate of return for a government bond with high
credit rating and a maturity similar to the time period over which the asset returns are

regressed. Rft is the time series of risk-free rates used in linear regression.

α: The return in excess of all the modelled risk parameters.

SMB: This is the average return on 3 small cap portfolios in excess of that from another 3
portfolios of large cap size (small minus big). The time series of values (SMBt) are used for
regression.

SMB =

1 1
(𝑆𝑚𝑎𝑙𝑙 𝑉𝑎𝑙𝑢𝑒 + 𝑆𝑚𝑎𝑙𝑙 𝑁𝑒𝑢𝑡𝑟𝑎𝑙 + 𝑆𝑚𝑎𝑙𝑙 𝐺𝑟𝑜𝑤𝑡ℎ) − (𝐵𝑖𝑔 𝑉𝑎𝑙𝑢𝑒 + 𝐵𝑖𝑔 𝑁𝑒𝑢𝑡𝑟𝑎𝑙 +
3 3

𝐵𝑖𝑔 𝐺𝑟𝑜𝑤𝑡ℎ)

HML: This is the difference in return of 2 value portfolios and 2 growth portfolios. This
investment style factor is added since value portfolios behave differently than growth
portfolios. Time series of the HMLt is used in the regression.
1 1
𝐻𝑀𝐿 = (𝑆𝑚𝑎𝑙𝑙 𝑉𝑎𝑙𝑢𝑒 + 𝐵𝑖𝑔 𝑉𝑎𝑙𝑢𝑒) − (𝑆𝑚𝑎𝑙𝑙 𝐺𝑟𝑜𝑤𝑡ℎ + 𝐵𝑖𝑔 𝐺𝑟𝑜𝑤𝑡ℎ)
2 2

Km: Market Return from a benchmark here is considered to be of a style and size neutral

benchmark relevant to the asset. Kmt - Rft is return in excess of the risk free rate at a time

t. All returns are calculated over the same period as the period used for rest of the
risk/return values.

4. Explain in words how the model improves upon CAPM.

Addition of Relevant Regressors: The FF model includes some specific factors that can
impact the returns of asset classes. While CAPM, looks at the market portfolio as a
benchmark to regress with the returns of an asset, the real-world application has shown that
a single factor is not enough to address the variances in the asset returns. Assets, or
portfolios managed according to a particular style have a more direct co-relation with a
benchmark of the same style (Value, Growth). FF model includes those factors that have a
specific impact
Real World Application: FF model includes the risk factors that are actually used in the asset
valuation by the investors. An asset with a lower market capitalization behaves differently
than an asset with a higher value. Higher cap assets have a higher liquidity than small cap
assets. SMB provides addition of this size factor. Likewise, the addition of HML captures the
investment style that differentiates portfolios from a neutral investment style (market
oriented).

5. Formulate the Fama-French regression using your stock’s returns, all the Fama-French
factors, and the benchmark returns.

Import Data: Adobe Stock from Yahoo Finance. XLS used to calculate returns
over the last closing price (~daily lag) for 2019 prices. Returns over the
last closing price are appended to the FF factor CSV. Note: E(Ri) is Daily
Return for each observation. E.G. E(Ri) is Price of Adobe at 1/3/2019 /
Price at 1/2/2018 – 1.

FF_ADBE_2019 <- read_excel("D:/Training/MScFE/Econometrics/M2/FF_ADB

E_2019.xlsx",
+ col_types = c("date", "numeric", "numeric",
+ "numeric", "numeric", "date", "numeric",
+ "numeric", "numeric"))
> View(FF_ADBE_2019)

Remove the last entry, as stock price for Adobe was not available for 31st
Dec 2019

Truncate_FF_ADBE_2019 <- FF_ADBE_2019[-c(252),]

View(Truncate_FF_ADBE_2019) FF_LM_ADBE_2019 <-

Run Regression: 3 Independent Variables (Market Return, SMB, HML). Same

frequency as that of Asset Return (Ri). Note: Market returns, SMB and HML
are also daily returns.

FF_LM_ADBE_2019 <- lm(Truncate_FF_ADBE_2019$`E(Ri)- Rf` ~

Truncate_FF_ADBE_2019$`Mkt-RF` + Truncate_FF_ADBE_2019$SMB +
Truncate_FF_ADBE_2019$HML)

{r echo=TRUE} summary(FF_LM_ADBE_2019)

Call:

lm(formula = Truncate_FF_ADBE_2019$`E(Ri)- Rf` ~

Truncate_FF_ADBE_2019$`Mkt-RF` +

Truncate_FF_ADBE_2019$SMB + Truncate_FF_ADBE_2019$HML)

Residuals:
Min 1Q Median 3Q Max
-4.8170 -0.4902 -0.0650 0.4090 4.7021

Coefficients:
 Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.00323 0.06141 -0.053 0.958
Truncate_FF_ADBE_2019$`Mkt-RF` 1.25044 0.07762 16.111 < 2e-16
***
Truncate_FF_ADBE_2019$SMB -0.14890 0.13684 -1.088 0.278
Truncate_FF_ADBE_2019$HML -0.78157 0.10494 -7.448 1.58e-12
***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.9625 on 247 degrees of freedom

Multiple R-squared: 0.5951, Adjusted R-squared: 0.5902
F-statistic: 121 on 3 and 247 DF, p-value: < 2.2e-16

Based on above the FF

E(r) – Rft = -0.003 + 1.25 *(Km - Rft) – 0.15*(SMBt) – 0.78 *(HMLt)

6. What is the Greek letter used in front of each factor?

The Betas measure the sensitivity of the asset returns to the corresponding risk factor.

β1 *(Km - Rf): Here β1 (1.25) measures the sensitivity of the asset return to the
general market risk.

β2 *(SMB): Here β2 (-0.15)measures the sensitivity of the asset return to the liquidity
risk, which is represented by the size difference between a small cap and large cap.

β3 *(HML): Here β1 (-0.78) measures the sensitivity of the asset return to the
investment style.

7. Which model performed better?

Let us compare with the CAPM model. We use 1 independent variable (Mkt-RF) instead of 3 in FF.
FF CAPM
Adjusted R2 59% 49.5%
F-Statistic Significance Same/Similar Same/Similar
Residual Std Err Lower (0.9625) Higher (1.069)

Generally, the FF model seems to have done a better job than CAPM, for explaining the variance
of the Adobe asset returns.
However, the Beta coefficient for SMB needs more research. It has a lower t-statistic of around 1-
Sigma. The coefficient is more likely to be zero. However, this is not a rejection of the FF model
itself. FF model requires a more careful SMB portfolio, one that reflects the investment style risk
and that is used as a type of benchmark by investors in the market in which Adobe stock trades in.
Generally, the FF has a better explanatory power due to the inclusion of more risk factors.
Appendix:
R Code for CAPM Model for the same Asset

> CAPM_LM_ADBE_2019 <- lm(formula = Truncate_FF_ADBE_2019$`E(Ri)- Rf`

~ Truncate_FF_ADBE_2019$`Mkt-RF`, data = Truncate_FF_ADBE_2019)
> summary(CAPM_LM_ADBE_2019)

Call:
lm(formula = Truncate_FF_ADBE_2019$`E(Ri)- Rf` ~
Truncate_FF_ADBE_2019$`Mkt-RF`,
data = Truncate_FF_ADBE_2019)

Residuals:
Min 1Q Median 3Q Max
-4.6090 -0.5323 -0.0308 0.4846 4.7701

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.02286 0.06795 0.336 0.737
Truncate_FF_ADBE_2019$`Mkt-RF` 1.28613 0.08202 15.680 <2e-16
***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 1.069 on 249 degrees of freedom

Multiple R-squared: 0.4968, Adjusted R-squared: 0.4948
F-statistic: 245.9 on 1 and 249 DF, p-value: < 2.2e-16