Herron Topic 1 - Practice for Section 03

FINA 6333 for Spring 2024

Author

Richard Herron

1 Announcements

  1. DataCamp
    1. Data Manipulation with pandas due by Friday, 2/9, at 11:59 PM
    2. Joining Data with pandas due by Friday, 2/16, at 11:59 PM
    3. Earn 10,000 XP due by Friday, 3/15, at 11:59 PM
  2. I posted Project 1 to Canvas
    1. Slides and notebook due by Friday, 2/23, at 11:59 PM
    2. Keep joining teams and let me know if you need help

2 10-Minute Recap

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import pandas_datareader as pdr
import yfinance as yf
%precision 4
pd.options.display.float_format = '{:.4f}'.format
%config InlineBackend.figure_format = 'retina'

First, we can easily download data in Python with two packages:

  1. yfinance to download data from Yahoo! Finance
  2. pandas-datareader to download data from Ken French (and FRED and others)

Second, there are “simple returns” and “log returns”

  1. Simple returns are the returns that investors receive that we learned in FINA 6331 and FINA 6333: \(r_t = \frac{p_t + d_t - p_{t-1}}{p_{t-1}}\)
  2. Log returns are the log of one plus simple returns. Why do we use them?
    1. Log returns are additive, while simple returns are multiplicative. This additive property makes math really easy with log returns: \(\log(\prod_{t=0}^T (1 + r_t)) = \sum_{t=0}^T \log(1+r_t)\), so \(r_{0,T} = \prod_{t=0}^T (1 + r_t) - 1 = e^{\sum_{t=0}^T \log(1+r_t)} - 1\)
    2. Log returns are almost normally distributed

We will almost always use simple returns. The exeception is time-consuming calculations, which we will often do in log returns to save us time.

Third, we can easily calculate portfolio returns in pandas!

  1. returns.mean(axis=1) is the equally-weighted portfolio return for the stocks in returns, rebalanced at the same frequency as the returns in returns!
  2. returns.dot(weights) lets us use weights in array weights instead of equally-weighted, but is still rebalanced at the same frequency as the returns in returns!

3 Practice

3.1 Download all available daily price data for tickers TSLA, F, AAPL, AMZN, and META to data frame prices

tickers = 'TSLA F AAPL AMZN META'
prices = yf.download(tickers=tickers)

prices
[*********************100%%**********************]  5 of 5 completed
Adj Close Close ... Open Volume
AAPL AMZN F META TSLA AAPL AMZN F META TSLA ... AAPL AMZN F META TSLA AAPL AMZN F META TSLA
Date
1972-06-01 NaN NaN 0.2419 NaN NaN NaN NaN 2.1532 NaN NaN ... NaN NaN 0.0000 NaN NaN NaN NaN 1091238 NaN NaN
1972-06-02 NaN NaN 0.2414 NaN NaN NaN NaN 2.1492 NaN NaN ... NaN NaN 2.1532 NaN NaN NaN NaN 1174468 NaN NaN
1972-06-05 NaN NaN 0.2414 NaN NaN NaN NaN 2.1492 NaN NaN ... NaN NaN 2.1492 NaN NaN NaN NaN 5209582 NaN NaN
1972-06-06 NaN NaN 0.2387 NaN NaN NaN NaN 2.1248 NaN NaN ... NaN NaN 2.1492 NaN NaN NaN NaN 1424158 NaN NaN
1972-06-07 NaN NaN 0.2373 NaN NaN NaN NaN 2.1127 NaN NaN ... NaN NaN 2.1248 NaN NaN NaN NaN 675088 NaN NaN
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
2024-02-05 187.6800 170.3100 11.5900 459.4100 181.0600 187.6800 170.3100 11.5900 459.4100 181.0600 ... 188.1500 170.2000 12.0100 469.8800 184.2600 69668800.0000 55081300.0000 81026800 40832400.0000 134294400.0000
2024-02-06 189.3000 169.1500 12.0700 454.7200 185.1000 189.3000 169.1500 12.0700 454.7200 185.1000 ... 186.8600 169.3900 11.6400 464.0000 177.2100 43490800.0000 42505500.0000 98636800 21655200.0000 122676000.0000
2024-02-07 189.4100 170.5300 12.8000 469.5900 187.5800 189.4100 170.5300 12.8000 469.5900 187.5800 ... 190.6400 169.4800 12.7300 458.0000 188.1800 53439000.0000 47174100.0000 137224000 23066000.0000 111535200.0000
2024-02-08 188.3200 169.8400 12.8300 470.0000 189.5600 188.3200 169.8400 12.8300 470.0000 189.5600 ... 189.3900 169.6500 12.8700 468.3200 189.0000 40962000.0000 42316500.0000 68653900 18815100.0000 83034000.0000
2024-02-09 188.8500 174.4500 12.6800 468.1100 193.5700 188.8500 174.4500 12.6800 468.1100 193.5700 ... 188.6500 170.9000 12.8100 472.9500 190.1800 43728065.0000 56886544.0000 46925766 18159624.0000 84009121.0000

13035 rows × 30 columns

3.2 Calculate all available daily returns and save to data frame returns

returns = (
    prices['Adj Close'] # slice adj close
    .iloc[:-1] # drop the last price because it might be intraday (i.e., not a close)
    .pct_change() # calculate simple returns
)

returns
AAPL AMZN F META TSLA
Date
1972-06-01 NaN NaN NaN NaN NaN
1972-06-02 NaN NaN -0.0019 NaN NaN
1972-06-05 NaN NaN 0.0000 NaN NaN
1972-06-06 NaN NaN -0.0113 NaN NaN
1972-06-07 NaN NaN -0.0057 NaN NaN
... ... ... ... ... ...
2024-02-02 -0.0054 0.0787 0.0033 0.2032 -0.0050
2024-02-05 0.0098 -0.0087 -0.0453 -0.0328 -0.0365
2024-02-06 0.0086 -0.0068 0.0414 -0.0102 0.0223
2024-02-07 0.0006 0.0082 0.0605 0.0327 0.0134
2024-02-08 -0.0058 -0.0040 0.0023 0.0009 0.0106

13034 rows × 5 columns

3.3 Slices returns for the 2020s and assign to returns_2020s

returns_2020s = returns.loc['2020':] # always use an unambiguos date format, like YYYY-MM-DD

returns_2020s
AAPL AMZN F META TSLA
Date
2020-01-02 0.0228 0.0272 0.0129 0.0221 0.0285
2020-01-03 -0.0097 -0.0121 -0.0223 -0.0053 0.0296
2020-01-06 0.0080 0.0149 -0.0054 0.0188 0.0193
2020-01-07 -0.0047 0.0021 0.0098 0.0022 0.0388
2020-01-08 0.0161 -0.0078 0.0000 0.0101 0.0492
... ... ... ... ... ...
2024-02-02 -0.0054 0.0787 0.0033 0.2032 -0.0050
2024-02-05 0.0098 -0.0087 -0.0453 -0.0328 -0.0365
2024-02-06 0.0086 -0.0068 0.0414 -0.0102 0.0223
2024-02-07 0.0006 0.0082 0.0605 0.0327 0.0134
2024-02-08 -0.0058 -0.0040 0.0023 0.0009 0.0106

1033 rows × 5 columns

3.4 Download all available data for the Fama and French daily benchmark factors to dictionary ff_all

I often use the following code snippet to find the exact name for the the daily benchmark factors file.

pdr.famafrench.get_available_datasets()[:5]
['F-F_Research_Data_Factors',
 'F-F_Research_Data_Factors_weekly',
 'F-F_Research_Data_Factors_daily',
 'F-F_Research_Data_5_Factors_2x3',
 'F-F_Research_Data_5_Factors_2x3_daily']
ff_all = pdr.DataReader(
    name='F-F_Research_Data_Factors_daily',
    data_source='famafrench',
    start='1900'
)
C:\Users\r.herron\AppData\Local\Temp\ipykernel_27592\2526882917.py:1: FutureWarning: The argument 'date_parser' is deprecated and will be removed in a future version. Please use 'date_format' instead, or read your data in as 'object' dtype and then call 'to_datetime'.
  ff_all = pdr.DataReader(

The DESCR key in the dictionary tells us about the data frames that pandas-datareader returns.

print(ff_all['DESCR'])
F-F Research Data Factors daily
-------------------------------

This file was created by CMPT_ME_BEME_RETS_DAILY using the 202312 CRSP database. The Tbill return is the simple daily rate that, over the number of trading days in the month, compounds to 1-month TBill rate from Ibbotson and Associates Inc. Copyright 2023 Kenneth R. French

  0 : (25649 rows x 4 cols)

3.5 Slice the daily benchmark factors, convert them to decimal returns, and assign to ff

ff = ff_all[0].div(100)

ff
Mkt-RF SMB HML RF
Date
1926-07-01 0.0010 -0.0025 -0.0027 0.0001
1926-07-02 0.0045 -0.0033 -0.0006 0.0001
1926-07-06 0.0017 0.0030 -0.0039 0.0001
1926-07-07 0.0009 -0.0058 0.0002 0.0001
1926-07-08 0.0021 -0.0038 0.0019 0.0001
... ... ... ... ...
2023-12-22 0.0021 0.0064 0.0009 0.0002
2023-12-26 0.0048 0.0069 0.0046 0.0002
2023-12-27 0.0016 0.0014 0.0012 0.0002
2023-12-28 -0.0001 -0.0036 0.0003 0.0002
2023-12-29 -0.0043 -0.0112 -0.0037 0.0002

25649 rows × 4 columns

3.6 Use the .cumprod() method to plot cumulative returns for these stocks in the 2020s

We use the .prod() method to calculate total returns, because \(r_{total} = r_{0,T} = \left[ \prod_{t=0}^T (1 + r_t) \right] -1\).

(
    returns_2020s # returns during the 2020s
    .add(1) # add 1 before we compound
    .prod() # compound all returns
    .sub(1) # subtract 1 to recover total returns
)
AAPL   1.6331
AMZN   0.8383
F      0.6090
META   1.2899
TSLA   5.7970
dtype: float64
cumret_cumprod =  (
    returns_2020s
    .add(1)
    .cumprod()
    .sub(1)
)
cumret_cumprod.mul(100).plot()

# https://stackoverflow.com/questions/25973581/how-to-format-axis-number-format-to-thousands-with-a-comma
from matplotlib import ticker
plt.gca().get_yaxis().set_major_formatter(ticker.FuncFormatter(lambda x, p: format(int(x), ',')))

plt.ylabel('Cumulative Return (%)')
plt.title(f'Cumulative Returns\n from {returns_2020s.index.min():%b %Y} to {returns_2020s.index.max():%b %Y}')
plt.show()

3.7 Use the .cumsum() method with log returns to plot cumulative returns for these stocks in the 2020s

cumret_cumsum =  (
    returns_2020s # start with simple returns
    .pipe(np.log1p) # convert to log returns
    .cumsum() # log returns are additive
    .pipe(np.expm1) # convert back to simple returns
)
np.allclose(cumret_cumsum, cumret_cumprod)
True
np.allclose(1*10**-9, 1.1*10**-9)
True
cumret_cumsum.mul(100).plot()

# https://stackoverflow.com/questions/25973581/how-to-format-axis-number-format-to-thousands-with-a-comma
from matplotlib import ticker
plt.gca().get_yaxis().set_major_formatter(ticker.FuncFormatter(lambda x, p: format(int(x), ',')))

plt.ylabel('Cumulative Return (%)')
plt.title(f'Cumulative Returns\n from {returns_2020s.index.min():%b %Y} to {returns_2020s.index.max():%b %Y}')
plt.show()

3.8 Use price data only to plot cumulative returns for these stocks in the 2020s

We can also calculate cumulative returns as the ratio of adjusted closes. That is \(R_{0,T} = \frac{AC_T}{AC_0} - 1\). The trick here is that \(FV_t = PV (1+r)^t\), so \((1+r)^t = \frac{FV_t}{PV}\).

returns_2020s.iloc[0]
AAPL   0.0228
AMZN   0.0272
F      0.0129
META   0.0221
TSLA   0.0285
Name: 2020-01-02 00:00:00, dtype: float64
cumret_prices = prices['Adj Close'].loc['2020':].iloc[:-1] / prices['Adj Close'].loc['2019'].iloc[-1] - 1
np.allclose(cumret_prices, cumret_cumprod)
True
cumret_prices.mul(100).plot()

# https://stackoverflow.com/questions/25973581/how-to-format-axis-number-format-to-thousands-with-a-comma
from matplotlib import ticker
plt.gca().get_yaxis().set_major_formatter(ticker.FuncFormatter(lambda x, p: format(int(x), ',')))

plt.ylabel('Cumulative Return (%)')
plt.title(f'Cumulative Returns\n from {returns_2020s.index.min():%b %Y} to {returns_2020s.index.max():%b %Y}')
plt.show()

3.9 Calculate the Sharpe Ratio for TSLA

Calculate the Sharpe Ratio with all available returns and 2020s returns. Recall the Sharpe Ratio is \(\frac{\overline{r_i - r_f}}{\sigma_i}\), where \(\sigma_i\) is the volatility of excess returns.

I suggest you write a function named calc_sharpe() to use for the rest of this notebook.

def calc_sharpe(ri, rf=ff['RF'], ppy=252):
    ri_rf = ri.sub(rf).dropna()
    return np.sqrt(ppy) * ri_rf.mean() / ri_rf.std()
calc_sharpe(ri=returns['TSLA'])
0.9261
calc_sharpe(ri=returns_2020s['TSLA'])
1.1212

Or, we can use the .pipe() method because ri is the first argument to calc_sharpe()!

returns['TSLA'].pipe(calc_sharpe)
0.9261
returns_2020s['TSLA'].pipe(calc_sharpe)
1.1212

3.10 Calculate the market beta for TSLA

Calculate the market beta with all available returns and 2020s returns. Recall we estimate market beta with the ordinary least squares (OLS) regression \(R_i-R_f = \alpha + \beta (R_m-R_f) + \epsilon\). We can estimate market beta with the covariance formula (i.e., \(\beta_i = \frac{Cov(R_i - R_f, R_m - R_f)}{Var(R_m-R_f)}\)) above for a univariate regression if we do not need goodness of fit statistics.

I suggest you write a function named calc_beta() to use for the rest of this notebook.

def calc_beta(ri, rf=ff['RF'], rm_rf=ff['Mkt-RF']):
    ri_rf = ri.sub(rf).dropna()
    return ri_rf.cov(rm_rf) / rm_rf.loc[ri_rf.index].var()
calc_beta(ri=returns['TSLA'])
1.4417
calc_beta(ri=returns_2020s['TSLA'])
1.5780

Or, we can use the .pipe() method because ri is the first argument to calc_sharpe()!

returns['TSLA'].pipe(calc_beta)
1.4417
returns_2020s['TSLA'].pipe(calc_beta)
1.5780

3.11 Guess the Sharpe Ratios for these stocks in the 2020s

3.12 Guess the market betas for these stocks in the 2020s

3.13 Calculate the Sharpe Ratios for these stocks in the 2020s

How good were your guesses?

for i in returns_2020s:
    sharpe_i = returns_2020s[i].pipe(calc_sharpe)
    print(f'Sharpe Ratio for {i}:\t {sharpe_i:0.2f}')
Sharpe Ratio for AAPL:   0.86
Sharpe Ratio for AMZN:   0.48
Sharpe Ratio for F:  0.42
Sharpe Ratio for META:   0.49
Sharpe Ratio for TSLA:   1.12

We can also use pandas notation to vectorize this calculation. First calculate excess returns as \(r_i - r_f\).

returns_2020s_excess = returns_2020s.sub(ff['RF'], axis=0).dropna()

Then use pandas notation to calculate means, standard deviations, and annualize.

(
    returns_2020s_excess
    .mean()
    .div(returns_2020s_excess.std())
    .mul(np.sqrt(252))
)
AAPL   0.8576
AMZN   0.4750
F      0.4240
META   0.4949
TSLA   1.1212
dtype: float64

Note: In a few weeks we will learn the .apply() method, which avoids the loop syntax.

returns_2020s.apply(calc_sharpe)
AAPL   0.8576
AMZN   0.4750
F      0.4240
META   0.4949
TSLA   1.1212
dtype: float64

3.14 Calculate the market betas for these stocks in the 2020s

How good were your guesses?

for i in returns_2020s:
    beta_i = returns_2020s[i].pipe(calc_beta)
    print(f'Beta for {i}:\t {beta_i:0.2f}')
Beta for AAPL:   1.15
Beta for AMZN:   1.04
Beta for F:  1.22
Beta for META:   1.27
Beta for TSLA:   1.58

Or we can follow out approach above to vectorize this calculation. First, we need to add a market excess return column to returns_2020s_excess.

returns_2020s_excess['Mkt-RF'] = ff['Mkt-RF']
returns_2020s_excess.head()
AAPL AMZN F META TSLA Mkt-RF
Date
2020-01-02 0.0228 0.0271 0.0128 0.0220 0.0285 0.0086
2020-01-03 -0.0098 -0.0122 -0.0224 -0.0054 0.0296 -0.0067
2020-01-06 0.0079 0.0148 -0.0055 0.0188 0.0192 0.0036
2020-01-07 -0.0048 0.0020 0.0098 0.0021 0.0387 -0.0019
2020-01-08 0.0160 -0.0079 -0.0001 0.0101 0.0491 0.0047 
vcv = returns_2020s_excess.cov()
vcv
AAPL AMZN F META TSLA Mkt-RF
AAPL 0.0004 0.0003 0.0002 0.0004 0.0005 0.0003
AMZN 0.0003 0.0006 0.0002 0.0004 0.0005 0.0002
F 0.0002 0.0002 0.0009 0.0003 0.0005 0.0003
META 0.0004 0.0004 0.0003 0.0009 0.0005 0.0003
TSLA 0.0005 0.0005 0.0005 0.0005 0.0018 0.0003
Mkt-RF 0.0003 0.0002 0.0003 0.0003 0.0003 0.0002 
vcv['Mkt-RF'].div(vcv.loc['Mkt-RF', 'Mkt-RF']).plot(kind='bar')
plt.xlabel('Ticker')
plt.ylabel('CAPM Beta')
plt.title('CAPM Betas')
plt.show()

Note: In a few weeks we will learn the .apply() method, which avoids the loop syntax.

returns_2020s.apply(calc_beta)
AAPL   1.1541
AMZN   1.0429
F      1.2231
META   1.2710
TSLA   1.5780
dtype: float64

3.15 Calculate the Sharpe Ratio for an equally weighted portfolio of these stocks in the 2020s

What do you notice?

returns_2020s.mean(axis=1).pipe(calc_sharpe)
0.9573

Because diversification reduces portfolio standard deviation less than the sum of its parts, the Sharpe Ratio of the equally weighted portfolio is less than the equally weighted mean of the single-stock Sharpe Ratios.

returns_2020s.apply(calc_sharpe).mean()
0.6745

3.16 Calculate the market beta for an equally weighted portfolio of these stocks in the 2020s

What do you notice?

Beta measures nondiversifiable risk, so \(\beta_P = \sum w_i \beta_i\)!

returns_2020s.mean(axis=1).pipe(calc_beta)
1.2538
returns_2020s.apply(calc_beta).mean()
1.2538

3.17 Calculate the market betas for these stocks every calendar year for every possible year

Save these market betas to data frame betas. Our current Python knowledge limits us to a for-loop, but we will learn easier and faster approaches soon!

betas = pd.DataFrame(
    index=range(1972, 2024),
    columns=returns.columns
)

betas.columns.name = 'Ticker'
betas.index.name = 'Year'

betas.tail()
Ticker AAPL AMZN F META TSLA
Year
2019 NaN NaN NaN NaN NaN
2020 NaN NaN NaN NaN NaN
2021 NaN NaN NaN NaN NaN
2022 NaN NaN NaN NaN NaN
2023 NaN NaN NaN NaN NaN 
for i in betas.index: 
    for c in betas.columns:
        betas.at[i, c] = returns.loc[str(i), c].pipe(calc_beta)

betas.tail()
Ticker AAPL AMZN F META TSLA
Year
2019 1.4751 1.2752 1.0733 1.2094 1.3262
2020 1.1174 0.6866 1.1052 0.9913 1.3041
2021 1.1957 0.9822 1.2396 1.2014 1.9891
2022 1.2386 1.5922 1.3824 1.6843 1.7414
2023 1.0369 1.4649 1.3947 1.6630 2.2218

3.18 Plot the time series of market betas

betas.plot()
plt.ylabel('CAPM Beta')
plt.title('CAPM Betas')
plt.show()