Appendix B — SQL primer

This brief appendix aims to serve two groups of users. The first group comprises those who have followed the material in the main part of the book and would like a quick introduction to SQL. The second group comprises those who know SQL (say, SAS’s PROC SQL) would like a quick introduction to the dplyr-based approach to R that we use in this book.

B.1 What are SQL and `dplyr`?

SQL is a specialized language for manipulating and retrieving tabular data used by almost all modern database systems.

The R package dplyr is a core part of the Tidyverse and perhaps the package we use the most in this book. From the Tidyverse website:

dplyr is a grammar of data manipulation, providing a consistent set of verbs that help you solve the most common data manipulation challenges:

mutate() adds new variables that are functions of existing variables

select() picks variables based on their names

filter() picks cases based on their values

summarize() reduces multiple values down to a single summary

arrange() changes the ordering of the rows

Table B.1: SQL translations of key dplyr verbs

`dplyr` verb (R)	SQL equivalent
`\|>`	`FROM`
`select()`	`SELECT`
`filter()`	`WHERE`
`group_by()`	`GROUP BY`
`arrange()`	`ORDER BY`
`mutate()`	used-defined columns
`summarize()`	used-defined aggregate columns

Prior to the advent of the dplyr in 2014, most users of R would have used base R functions and operators, such as subset(), $, and [. However, dplyr provides a much more consistent framework for manipulating data that is easier to learn, especially for users familiar with SQL. This is unsurprising given that SQL provides something like a “grammar of data manipulation” of its own. In fact, each dplyr verb has an SQL equivalent keyword or concept, as seen in Table B.1, which provides a translation table of sorts between dplyr and SQL.

PostgreSQL
parquet

In this appendix, we use the following packages:

library(DBI)
library(dplyr)

db <- dbConnect(RPostgres::Postgres())

crsp.dsf <- tbl(db, Id(table = "dsf", schema = "crsp"))
crsp.dsi <- tbl(db, Id(table = "dsi", schema = "crsp"))

In this appendix, we use the following packages:

library(DBI)
library(dplyr)
library(farr)

As in earlier chapters, we set up a database connection that we can use within R. We also create remote data frames for each of the two tables from WRDS that we use in this appendix: crsp.dsf and crsp.dsi. A wrinkle with our parquet-based approach to DuckDB is that to access the tables from SQL, we need to create named tables that refer to the underlying parquet files. The create_view() function does this. Note that we use CREATE VIEW rather than CREATE TABLE so that the data are not read into memory (this is costly with crsp.dsf).

create_view <- function(conn, table, schema = "", 
                         data_dir = Sys.getenv("DATA_DIR"))  {
  dbExecute(conn, paste0("CREATE SCHEMA IF NOT EXISTS ", schema))
  file_path <- file.path(data_dir, schema, paste0(table, ".parquet"))
  df_sql <- paste0("CREATE VIEW ", schema, ".", table, " AS ",
                   "SELECT * FROM read_parquet('", file_path, 
                   "')")
  DBI::dbExecute(conn, dplyr::sql(df_sql))
}

db <- dbConnect(duckdb::duckdb())

crsp.dsi <- create_view(db, table = "dsi", schema = "crsp") 
crsp.dsf <- create_view(db, table = "dsf", schema = "crsp")

B.2 SQL terms `SELECT` and `FROM`

Let’s begin with a basic SQL query. (Note that, throughout this appendix, we only display the first five records when there are more than five records returned by a query. Note that a record is SQL terminology for what we might call a row in the context of an R data frame.)

SELECT date, vwretd, ewretd
FROM crsp.dsi

date	vwretd	ewretd
1925-12-31
1926-01-02	0.005689	0.009516
1926-01-04	0.000706	0.005780
1926-01-05	-0.004821	-0.001927
1926-01-06	-0.000423	0.001182

The query above

extracts the data in three columns (date, vwretd, ewretd) (the first line)
from the table named crsp.dsi (the second line)

While the syntax differs, SQL’s SELECT operates very much like select() from dplyr.

Translating this into dplyr code using the pipe operator (|>), it’s easy to see that the order of presentation is one of the big differences between SQL and dplyr.

To use dplyr we first need to set up the table on which to operate; once we’ve done so we can see that SQL’s FROM is implicit in the |> operator.

crsp.dsi |>
  select(date, vwretd, ewretd)

date	vwretd	ewretd
1925-12-31
1926-01-02	0.005689	0.009516
1926-01-04	0.000706	0.005780
1926-01-05	-0.004821	-0.001927
1926-01-06	-0.000423	0.001182

B.3 SQL `WHERE`

The filter() verb from dplyr corresponds to WHERE in SQL. Note that WHERE goes after the FROM clause in SQL, though in practice the query optimizer will execute the filter implied by the WHERE clause before executing other elements of a query (it would be wasteful to perform calculations on data that are going to be filtered out later on).

SELECT date, vwretd, ewretd
FROM crsp.dsi
WHERE date = '2015-01-02'

date	vwretd	ewretd
2015-01-02	-0.000242	0.001470

In dplyr, this query could be written as follows:

crsp.dsi |>
  select(date, vwretd, ewretd) |>
  filter(date == '2015-01-02')

date	vwretd	ewretd
2015-01-02	-0.000242	0.001470

But it could also be written with the filter coming first.

crsp.dsi |>
  filter(date == '2015-01-02') |>
  select(date, vwretd, ewretd)

date	vwretd	ewretd
2015-01-02	-0.000242	0.001470

Note that each of these dplyr queries is implemented in the same way by the PostgreSQL query optimizer, as can be seen by examining the output from explain().

crsp.dsi |>
  select(date, vwretd, ewretd) |>
  filter(date == '2015-01-02') |>
  explain()

<SQL>
SELECT "date", "vwretd", "ewretd"
FROM "crsp"."dsi"
WHERE ("date" = '2015-01-02')

<PLAN>
                                                               QUERY PLAN
1 Index Scan using dsi_date_idx on dsi  (cost=0.29..8.30 rows=1 width=20)
2                                 Index Cond: (date = '2015-01-02'::date)

crsp.dsi |>
  filter(date == '2015-01-02') |>
  select(date, vwretd, ewretd) |>
  explain()

<SQL>
SELECT "date", "vwretd", "ewretd"
FROM "crsp"."dsi"
WHERE ("date" = '2015-01-02')

<PLAN>
                                                               QUERY PLAN
1 Index Scan using dsi_date_idx on dsi  (cost=0.29..8.30 rows=1 width=20)
2                                 Index Cond: (date = '2015-01-02'::date)

By adding EXPLAIN to the SQL above, we can see also that the dplyr queries are viewed by the query optimizer as equivalent to that SQL.

EXPLAIN
SELECT date, vwretd, ewretd
FROM crsp.dsi
WHERE date = '2015-01-02'

                                                               QUERY PLAN
1 Index Scan using dsi_date_idx on dsi  (cost=0.29..8.30 rows=1 width=20)
2                                 Index Cond: (date = '2015-01-02'::date)

This illustrates a nice feature of SQL engines, namely that you can leave some of the details of the query to the software and focus on the higher-level requirements of your query (i.e., which observations to keep, etc.).

B.4 SQL `ORDER BY`

An R data frame can be considered a list of vectors, where each vector has a well-defined order. In contrast, SQL tables should be not considered to have any particular order when there is no ORDER BY clause.¹

An ORDER BY clause is placed at the end of an SQL query (reflecting in this case the order of operations) followed by the columns that the query is to be sorted on.

SELECT date, vwretd, ewretd
FROM crsp.dsi
ORDER BY vwretd

date	vwretd	ewretd
1987-10-19	-0.171346	-0.103897
1929-10-29	-0.119539	-0.134856
2020-03-16	-0.118168	-0.107631
1929-10-28	-0.113022	-0.095202
1929-11-06	-0.097020	-0.083819

Thus, ORDER BY works very much like arrange().

crsp.dsi |>
  select(date, vwretd, ewretd) |>
  arrange(vwretd)

date	vwretd	ewretd
1987-10-19	-0.171346	-0.103897
1929-10-29	-0.119539	-0.134856
2020-03-16	-0.118168	-0.107631
1929-10-28	-0.113022	-0.095202
1929-11-06	-0.097020	-0.083819

To reverse the order, use the DESC keyword after the relevant variable.

SELECT date, vwretd, ewretd
FROM crsp.dsi
ORDER BY vwretd DESC, date

date	vwretd	ewretd
1925-12-31
1933-03-15	0.156838	0.203020
1929-10-30	0.122155	0.125714
2008-10-13	0.114918	0.107422
1931-10-06	0.111278	0.088742

Thus, DESC is very much equivalent to desc() in dplyr apart from the minor syntactical difference that desc() is written as a function of the relevant variable.

crsp.dsi |>
  select(date, vwretd, ewretd) |>
  arrange(desc(vwretd), date)

date	vwretd	ewretd
1925-12-31
1933-03-15	0.156838	0.203020
1929-10-30	0.122155	0.125714
2008-10-13	0.114918	0.107422
1931-10-06	0.111278	0.088742

It is important to note that some differences in the implementation details between data frames in R and tables in SQL are seen with arrange(). While a data frame in R is fundamentally a list of ordered vectors, SQL tables are best thought of as sets of rows without a natural order. As such, one needs to pay attention to when arrange() is implemented in a query that depends on it.

To illustrate this issue, suppose we get stock returns for Apple and order them from largest to smallest returns.²

apple <-
  crsp.dsf |>
  filter(permno == 14593L) |>
  select(permno, date, ret) 

apple |>
  arrange(desc(ret))

permno	date	ret
14593	1980-12-12
14593	1997-08-06	0.332278
14593	1998-01-02	0.238095
14593	1996-07-18	0.237037
14593	1998-01-06	0.192913

The following code illustrates that the order created above is lost when we merge with crsp.dsi. If you are using window queries with SQL data sources, use window_order() to order data within windows.

apple |>
  inner_join(crsp.dsi, by = "date") |>
  select(permno, date, ret, vwretd)

permno	date	ret	vwretd
14593	1980-12-12		0.014859
14593	1980-12-15	-0.052061	0.001605
14593	1980-12-16	-0.073227	0.007485
14593	1980-12-17	0.024691	0.016168
14593	1980-12-18	0.028916	0.004134

B.5 SQL approach to `mutate()`

Creating new variables in SQL is quite straightforward. In addition to basic mathematical operators, such as + and *, PostgreSQL and DuckDB have an extensive array of functions available for use in queries.³

The names of calculated variables can be specified using AS, as in the following query:

SELECT date, 
  vwretd - ewretd AS ret_diff,
  ln(1 + vwretd) AS log_vwretd,
  date_part('year', date) AS year
FROM crsp.dsi

date	ret_diff	log_vwretd	year
1925-12-31			1925
1926-01-02	-0.003827	0.005673	1926
1926-01-04	-0.005074	0.000706	1926
1926-01-05	-0.002894	-0.004833	1926
1926-01-06	-0.001605	-0.000423	1926

The translation of this query into dplyr requires the use of the mutate() verb, but this translation is straightforward, as can be seen below:

crsp.dsi |>
  mutate(ret_diff = vwretd - ewretd, 
         log_vwretd = log(1 + vwretd),
         year = year(date)) |>
  select(date, ret_diff, log_vwretd, year)

date	ret_diff	log_vwretd	year
1925-12-31			1925
1926-01-02	-0.003827	0.005673	1926
1926-01-04	-0.005074	0.000706	1926
1926-01-05	-0.002894	-0.004833	1926
1926-01-06	-0.001605	-0.000423	1926

We see that the dplyr code is slightly more verbose than the SQL because the SQL equivalent of mutate() is an implicit part of the SELECT portion of the statement.

Note that we use the ln() in SQL, but use log() in the dplyr code. In SQL, log() returns the base-10 logarithm, but dplyr will translate common functions such as log() to their SQL equivalents.⁴ An alternative approach would be to use ln() with dplyr. Because ln() is not a recognized R function, ln() is passed to SQL unmodified. Similarly, year() is translated by dbplyr to its nearest equivalent in SQL (for year() something using SQL’s EXTRACT that is equivalent to date_part()). We can use show_query() to see the generated SQL.

crsp.dsi |>
  mutate(ret_diff = vwretd - ewretd, 
         log_vwretd = log(1 + vwretd),
         year = year(date)) |>
  select(date, ret_diff, log_vwretd, year) |>
  show_query()

<SQL>
SELECT
  "date",
  "vwretd" - "ewretd" AS "ret_diff",
  LN(1.0 + "vwretd") AS "log_vwretd",
  EXTRACT(year FROM "date") AS "year"
FROM "crsp"."dsi"

B.6 SQL `GROUP BY` and aggregates

One use for group_by() in R is to create summary statistics and the like for each of the groups implied by the group_by() variables. In SQL, GROUP BY is used with aggregate functions, which are like the functions used with summarize().

Suppose we want to get the maximum and minimum returns, as well as a measure of the volatility of the returns, for each year. In SQL, we could do the following:

SELECT date_part('year', date) AS year,
  max(vwretd) AS max_ret,
  min(vwretd) AS min_ret,
  stddev(ln(1 + vwretd)) AS sd_ret
FROM crsp.dsi
GROUP BY year
ORDER BY sd_ret DESC

year	max_ret	min_ret	sd_ret
1925
1932	0.109846	-0.073118	0.027422
1933	0.156838	-0.093272	0.025666
2008	0.114918	-0.089818	0.025397
2020	0.091556	-0.118168	0.021380

And the dplyr equivalent would look like this:

crsp.dsi |>
  mutate(year = year(date)) |>
  group_by(year) |>
  summarize(max_ret = max(vwretd, na.rm = TRUE),
            min_ret = min(vwretd, na.rm = TRUE),
            sd_ret = sd(log(1 + vwretd), na.rm = TRUE)) |>
  arrange(desc(sd_ret))

year	max_ret	min_ret	sd_ret
1925
1932	0.109846	-0.073118	0.027422
1933	0.156838	-0.093272	0.025666
2008	0.114918	-0.089818	0.025397
2020	0.091556	-0.118168	0.021380

Note that na.rm = TRUE could be omitted, as SQL always omits these values. The inclusion of this argument serves only as a reminder to the user of the code that NA values are omitted.⁵

See https://stackoverflow.com/questions/20050341 on this point.↩︎
Note that the NA value from Apple’s first day of trading is viewed as being the largest value due to the way that SQL collates NULL values.↩︎
See https://www.postgresql.org/docs/current/functions.html.↩︎
See https://www.postgresql.org/docs/current/functions-math.html.↩︎
Setting na.rm = TRUE also suppresses warnings from dplyr regarding the handling of missing values in SQL.↩︎

B.1 What are SQL and dplyr?

B.2 SQL terms SELECT and FROM

B.3 SQL WHERE

B.4 SQL ORDER BY

B.5 SQL approach to mutate()

B.6 SQL GROUP BY and aggregates