tidyfinance
packagetidyfinance
is an R
package on
CRAN that contains a set of helper functions for empirical research
in financial economics, addressing a variety of topics covered in Tidy Finance with R (TFWR).
We designed the package to provide easy shortcuts for the applications
that we discuss in the book. If you want to inspect the details of the
package or propose new features, feel free to visit the package
repository on Github.
In this vignette, we demonstrate the features of the initial release. We decided to focus on functions that allow you to download the data that we use in TFWR.
You can install the released version of tidyfinance
from
CRAN via:
You can install the development version of tidyfinance from GitHub
using the pak
package:
Let’s start by loading the package
The main function is
download_data(type, start_date, end_date)
with supported
type:
list_supported_types()
#> # A tibble: 317 × 3
#> type dataset_name domain
#> <chr> <chr> <chr>
#> 1 factors_q5_daily q5_factors_daily_2023.csv Global Q
#> 2 factors_q5_weekly q5_factors_weekly_2023.csv Global Q
#> 3 factors_q5_weekly_w2w q5_factors_weekly_w2w_2023.csv Global Q
#> 4 factors_q5_monthly q5_factors_monthly_2023.csv Global Q
#> 5 factors_q5_quarterly q5_factors_quarterly_2023.csv Global Q
#> 6 factors_q5_annual q5_factors_annual_2023.csv Global Q
#> 7 factors_ff_3_monthly Fama/French 3 Factors Fama-French
#> 8 factors_ff_3_weekly Fama/French 3 Factors [Weekly] Fama-French
#> 9 factors_ff_3_daily Fama/French 3 Factors [Daily] Fama-French
#> 10 factors_ff_5_2x3_monthly Fama/French 5 Factors (2x3) Fama-French
#> # ℹ 307 more rows
So, for instance, if you want to download monthly Fama-French Three-Factor data, you can call:
download_data("factors_ff_3_monthly", "2020-01-01", "2020-12-31")
#> # A tibble: 12 × 5
#> date mkt_excess smb hml risk_free
#> <date> <dbl> <dbl> <dbl> <dbl>
#> 1 2020-01-01 -0.0011 -0.0313 -0.0625 0.0013
#> 2 2020-02-01 -0.0813 0.0107 -0.038 0.0012
#> 3 2020-03-01 -0.134 -0.0479 -0.139 0.0013
#> 4 2020-04-01 0.136 0.0245 -0.0134 0
#> 5 2020-05-01 0.0558 0.0249 -0.0485 0.0001
#> 6 2020-06-01 0.0246 0.0269 -0.0223 0.0001
#> 7 2020-07-01 0.0577 -0.023 -0.0144 0.0001
#> 8 2020-08-01 0.0763 -0.0028 -0.0288 0.0001
#> 9 2020-09-01 -0.0363 -0.0003 -0.0265 0.0001
#> 10 2020-10-01 -0.021 0.0427 0.0431 0.0001
#> 11 2020-11-01 0.125 0.0572 0.0215 0.0001
#> 12 2020-12-01 0.0463 0.0479 -0.0134 0.0001
Under the hood, the function uses the frenchdata
package
(see its documentation here) and
applies some cleaning steps, as in TFWR. If you haven’t installed
frenchdata
yet, you’ll get prompted to install it first
before you can download this specific data type.
You can also access q-Factor data in this way, by calling:
download_data("factors_q5_daily", "2020-01-01", "2020-12-31")
#> # A tibble: 253 × 7
#> date risk_free mkt_excess me ia roe eg
#> <date> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 2020-01-02 0.000055 0.00863 -0.0112 -0.00170 0.000681 0.00338
#> 2 2020-01-03 0.000055 -0.00673 0.00234 -0.00192 -0.00156 0.000703
#> 3 2020-01-06 0.000055 0.00360 -0.00359 -0.00408 -0.00479 0.000552
#> 4 2020-01-07 0.000055 -0.00192 -0.00139 -0.00323 -0.00511 -0.00272
#> 5 2020-01-08 0.000055 0.00467 -0.00107 -0.00120 0.00454 0.00622
#> 6 2020-01-09 0.000055 0.00649 -0.00684 -0.000647 0.00294 0.00522
#> 7 2020-01-10 0.000055 -0.00335 -0.00236 -0.00205 0.0004 0.00338
#> 8 2020-01-13 0.000055 0.00729 -0.00199 0.00210 0.00305 0.000733
#> 9 2020-01-14 0.000055 -0.000559 0.00474 -0.00110 -0.00992 -0.00717
#> 10 2020-01-15 0.000055 0.00164 0.00305 -0.00265 -0.0011 0.00121
#> # ℹ 243 more rows
To ensure that we can extend the functionality of the download
functions for specific types, we also provide domain-specific download
functions. The download_data("factors_ff_3_monthly")
actually calls
download_data_factors("factors_ff_3_monthly")
, which in
turn calls
download_data_factors_ff("factors_ff_3_monthly")
. Why did
we decide to have these nested function approach?
Suppose that the q-Factor data changes its URL path and our original
function does not work anymore. In this case, you can replace the
default url
value in
download_data_factors_q(type, start_date, end_date, url)
to
apply the usual cleaning steps.
This feature becomes more apparent for other data sources such as
wrds_crsp_monthly
. Note that you need to have valid WRDS
credentials and need to set them correctly (check
?set_wrds_connection()
and WRDS,
CRSP, and Compustat in TFWR). If you want to download the standard
monthly CRSP data, you can call:
If you want to add further columns, you can add them via
additional_columns
to
download_data_wrds_crsp()
, for instance:
download_data_wrds_crsp("wrds_crsp_monthly", "2020-01-01", "2020-12-31", additional_columns = "mthvol")
Note that the function downloads CRSP v2 as default, as we do in our
book since February 2024. If you want to download the old version of
CRSP before the update, you can use the version = v1
parameter in download_data_wrds_crsp()
.
As another example, you can do the same for Compustat:
download_data_wrds_compustat("wrds_compustat_annual", "2000-01-01", "2020-12-31", additional_columns = c("acoxar", "amc", "aldo"))
Check out the list of supported types and the corresponding download functions for more information on the respective customization options. We decided to provide limited functionality for the initial release on purpose and rather respond to community request than overengineer the package from the start.
We include functions to check out content from TFWR in your browser. If you want to list all available R chapters, simply call the following function:
list_tidy_finance_chapters()
#> [1] "setting-up-your-environment"
#> [2] "introduction-to-tidy-finance"
#> [3] "accessing-and-managing-financial-data"
#> [4] "wrds-crsp-and-compustat"
#> [5] "trace-and-fisd"
#> [6] "other-data-providers"
#> [7] "beta-estimation"
#> [8] "univariate-portfolio-sorts"
#> [9] "size-sorts-and-p-hacking"
#> [10] "value-and-bivariate-sorts"
#> [11] "replicating-fama-and-french-factors"
#> [12] "fama-macbeth-regressions"
#> [13] "fixed-effects-and-clustered-standard-errors"
#> [14] "difference-in-differences"
#> [15] "factor-selection-via-machine-learning"
#> [16] "option-pricing-via-machine-learning"
#> [17] "parametric-portfolio-policies"
#> [18] "constrained-optimization-and-backtesting"
#> [19] "wrds-dummy-data"
#> [20] "cover-and-logo-design"
#> [21] "clean-enhanced-trace-with-r"
#> [22] "proofs"
#> [23] "hex-sticker"
#> [24] "changelog"
The function returns a character vector containing the names of the chapters available in TFWR. If you want to look at a specific chapter, you can call:
This opens either the specific chapter you requested or the main index page in your default web browser.
We discuss winsorization in TFWR, so we figured providing this function could be useful:
library(dplyr)
set.seed(123)
data <- tibble(x = rnorm(100)) |>
arrange(x)
data |>
mutate(x_winsorized = winsorize(x, 0.01))
#> # A tibble: 100 × 2
#> x x_winsorized
#> <dbl> <dbl>
#> 1 -2.31 -1.97
#> 2 -1.97 -1.97
#> 3 -1.69 -1.69
#> 4 -1.55 -1.55
#> 5 -1.27 -1.27
#> 6 -1.27 -1.27
#> 7 -1.22 -1.22
#> 8 -1.14 -1.14
#> 9 -1.12 -1.12
#> 10 -1.07 -1.07
#> # ℹ 90 more rows
If you rather want to replace the bottom and top quantiles of your
distribution with missing values, then you can use
trim()
data |>
mutate(x_trimmed = trim(x, 0.01))
#> # A tibble: 100 × 2
#> x x_trimmed
#> <dbl> <dbl>
#> 1 -2.31 NA
#> 2 -1.97 -1.97
#> 3 -1.69 -1.69
#> 4 -1.55 -1.55
#> 5 -1.27 -1.27
#> 6 -1.27 -1.27
#> 7 -1.22 -1.22
#> 8 -1.14 -1.14
#> 9 -1.12 -1.12
#> 10 -1.07 -1.07
#> # ℹ 90 more rows
We also discuss the importance of providing summary statistics of your data, so there is also a function for that:
create_summary_statistics(data, x, detail = TRUE)
#> # A tibble: 1 × 15
#> variable n mean sd min q01 q05 q10 q25 q50 q75 q90
#> <chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 x 100 0.0904 0.913 -2.31 -1.97 -1.27 -1.07 -0.494 0.0618 0.692 1.26
#> # ℹ 3 more variables: q95 <dbl>, q99 <dbl>, max <dbl>
We have two more experimental functions in the sense that it is
unclear in which direction they might evolve. First you can assign
portfolios based on a sorting variable using
assign_portfolio()
:
data <- tibble(
id = 1:100,
exchange = sample(c("NYSE", "NASDAQ"), 100, replace = TRUE),
market_cap = runif(100, 1e6, 1e9)
)
data |>
mutate(
portfolio = assign_portfolio(
pick(everything()), "market_cap", list(n_portfolios = 5, breakpoint_exchanges = "NYSE"))
)
#> # A tibble: 100 × 4
#> id exchange market_cap portfolio
#> <int> <chr> <dbl> <int>
#> 1 1 NASDAQ 784790691. 4
#> 2 2 NASDAQ 10420475. 1
#> 3 3 NASDAQ 779286817. 4
#> 4 4 NYSE 729661261. 4
#> 5 5 NASDAQ 630501721. 3
#> 6 6 NASDAQ 481429919. 3
#> 7 7 NASDAQ 157480215. 1
#> 8 8 NYSE 9207304. 1
#> 9 9 NASDAQ 453005936. 3
#> 10 10 NASDAQ 492801035. 3
#> # ℹ 90 more rows
Second, you can estimate the coefficients of a linear model specified
by one or more independent variable using
estimate_model()
:
We are curious to learn in which direction we should extend the
package, so please consider opening an issue in the package
repository. For instance, we could support more data sources, add
more parameters to the download_*
family of functions, or
we could put more emphasis on the generality of portfolio assignment or
other modeling functions.