R Quantitative Analysis Package Integrations in tidyquant

knitr::opts_chunk$set(message = FALSE,
                      warning = FALSE,
                      fig.width = 8, 
                      fig.height = 4.5,
                      fig.align = 'center',
                      out.width='95%')
# devtools::load_all() # Travis CI fails on load_all()

Functions that leverage the quantitative analysis functionality of xts, zoo, quantmod, TTR, and PerformanceAnalytics

Overview

There's a wide range of useful quantitative analysis functions that work with time-series objects. The problem is that many of these wonderful functions don't work with data frames or the tidyverse workflow. That is until now! The tidyquant package integrates the most useful functions from the xts, zoo, quantmod, TTR, and PerformanceAnalytics packages. This vignette focuses on the following core functions to demonstrate how the integratation works with the quantitative finance packages:

Refer to Performance Analysis with tidyquant for a full discussion on performance analysis and portfolio attribution with tidyquant.

Prerequisites

Load the tidyquant package to get started.

# Loads tidyquant, lubridate, xts, quantmod, TTR 
library(tidyverse)
library(tidyquant)

1.0 Function Compatibility

tq_transmute_fun_options() returns a list the compatible mutate functions by each package. We'll discuss these options by package briefly.

tq_transmute_fun_options() %>% str()

zoo Functionality

# Get zoo functions that work with tq_transmute and tq_mutate
tq_transmute_fun_options()$zoo

The zoo functions that are compatible are listed above. Generally speaking, these are the:

xts Functionality

# Get xts functions that work with tq_transmute and tq_mutate
tq_transmute_fun_options()$xts

The xts functions that are compatible are listed above. Generally speaking, these are the:

quantmod Functionality

# Get quantmod functions that work with tq_transmute and tq_mutate
tq_transmute_fun_options()$quantmod

The quantmod functions that are compatible are listed above. Generally speaking, these are the:

TTR Functionality

# Get TTR functions that work with tq_transmute and tq_mutate
tq_transmute_fun_options()$TTR

Here' a brief description of the most popular functions from TTR:

PerformanceAnalytics Functionality

# Get PerformanceAnalytics functions that work with tq_transmute and tq_mutate
tq_transmute_fun_options()$PerformanceAnalytics

The PerformanceAnalytics mutation functions all deal with returns:

2.0 Quantitative Power In Action

We'll go through some examples, but first let's get some data. The FANG data set will be used which consists of stock prices for FB, AMZN, NFLX, and GOOG from the beginning of 2013 to the end of 2016.

data("FANG")

FANG

Example 1: Use quantmod periodReturn to Convert Prices to Returns

The quantmod::periodReturn() function generates returns by periodicity. We'll go through a couple usage cases.

Example 1A: Getting and Charting Annual Returns

We want to use the adjusted closing prices column (adjusted for stock splits, which can make it appear that a stock is performing poorly if a split is included). We set select = adjusted. We research the periodReturn function, and we found that it accepts type = "arithmetic" and period = "yearly", which returns the annual returns.

FANG_annual_returns <- FANG %>%
    group_by(symbol) %>%
    tq_transmute(select     = adjusted, 
                 mutate_fun = periodReturn, 
                 period     = "yearly", 
                 type       = "arithmetic")
FANG_annual_returns

Charting annual returns is just a quick use of the ggplot2 package.

FANG_annual_returns %>%
    ggplot(aes(x = date, y = yearly.returns, fill = symbol)) +
    geom_col() +
    geom_hline(yintercept = 0, color = palette_light()[[1]]) +
    scale_y_continuous(labels = scales::percent) +
    labs(title = "FANG: Annual Returns",
         subtitle = "Get annual returns quickly with tq_transmute!",
         y = "Annual Returns", x = "") + 
    facet_wrap(~ symbol, ncol = 2, scales = "free_y") +
    theme_tq() + 
    scale_fill_tq()

Example 1B: Getting Daily Log Returns

Daily log returns follows a similar approach. Normally I go with a transmute function, tq_transmute, because the periodReturn function accepts different periodicity options, and anything other than daily will blow up a mutation. But, in our situation the period returns periodicity is the same as the stock prices periodicity (both daily), so we can use either. We want to use the adjusted closing prices column (adjusted for stock splits, which can make it appear that a stock is performing poorly if a split is included), so we set select = adjusted. We researched the periodReturn function, and we found that it accepts type = "log" and period = "daily", which returns the daily log returns.

FANG_daily_log_returns <- FANG %>%
    group_by(symbol) %>%
    tq_transmute(select     = adjusted, 
                 mutate_fun = periodReturn, 
                 period     = "daily", 
                 type       = "log",
                 col_rename = "monthly.returns")
FANG_daily_log_returns %>%
    ggplot(aes(x = monthly.returns, fill = symbol)) +
    geom_density(alpha = 0.5) +
    labs(title = "FANG: Charting the Daily Log Returns",
         x = "Monthly Returns", y = "Density") +
    theme_tq() +
    scale_fill_tq() + 
    facet_wrap(~ symbol, ncol = 2)

Example 2: Use xts to.period to Change the Periodicity from Daily to Monthly

The xts::to.period function is used for periodicity aggregation (converting from a lower level periodicity to a higher level such as minutes to hours or months to years). Because we are seeking a return structure that is on a different time scale than the input (daily versus weekly), we need to use a transmute function. We select tq_transmute() and pass the open, high, low, close and volume columns via select = open:volume. Looking at the documentation for to.period, we see that it accepts a period argument that we can set to "weeks". The result is the OHLCV data returned with the dates changed to one day per week.

FANG %>%
    group_by(symbol) %>%
    tq_transmute(select     = open:volume, 
                 mutate_fun = to.period, 
                 period     = "months")

A common usage case is to reduce the number of points to smooth time series plots. Let's check out difference between daily and monthly plots.

Without Periodicity Aggregation

FANG_daily <- FANG %>%
    group_by(symbol)

FANG_daily %>%
    ggplot(aes(x = date, y = adjusted, color = symbol)) +
    geom_line(size = 1) +
    labs(title = "Daily Stock Prices",
         x = "", y = "Adjusted Prices", color = "") +
    facet_wrap(~ symbol, ncol = 2, scales = "free_y") +
    scale_y_continuous(labels = scales::dollar) +
    theme_tq() + 
    scale_color_tq()

With Monthly Periodicity Aggregation

FANG_monthly <- FANG %>%
    group_by(symbol) %>%
    tq_transmute(select     = adjusted, 
                 mutate_fun = to.period, 
                 period     = "months")

FANG_monthly %>%
    ggplot(aes(x = date, y = adjusted, color = symbol)) +
    geom_line(size = 1) +
    labs(title = "Monthly Stock Prices",
         x = "", y = "Adjusted Prices", color = "") +
    facet_wrap(~ symbol, ncol = 2, scales = "free_y") +
    scale_y_continuous(labels = scales::dollar) +
    theme_tq() + 
    scale_color_tq()

Example 3: Use TTR runCor to Visualize Rolling Correlations of Returns

Return correlations are a common way to analyze how closely an asset or portfolio mimics a baseline index or fund. We will need a set of returns for both the stocks and baseline. The stock will be the FANG data set and the baseline will be the Spdr XLK technology sector. We have the prices for the "FANG" stocks, so we use tq_get to retrieve the "XLK" prices. The returns can be calculated from the "adjusted" prices using the process in Example 1.

# Asset Returns
FANG_returns_monthly <- FANG %>%
    group_by(symbol) %>%
    tq_transmute(select     = adjusted, 
                 mutate_fun = periodReturn,
                 period     = "monthly")

# Baseline Returns
baseline_returns_monthly <- "XLK" %>%
    tq_get(get  = "stock.prices",
           from = "2013-01-01", 
           to   = "2016-12-31") %>%
    tq_transmute(select     = adjusted, 
                 mutate_fun = periodReturn,
                 period     = "monthly")

Next, join the asset returns with the baseline returns by date.

returns_joined <- left_join(FANG_returns_monthly, 
                            baseline_returns_monthly,
                            by = "date")
returns_joined

The TTR::runCor function can be used to evaluate rolling correlations using the xy pattern. Looking at the documentation (?runCor), we can see that the arguments include x and y along with a few additional arguments including n for the width of the rolling correlation. Because the scale is monthly, we'll go with n = 6 for a 6-month rolling correlation. The col_rename argument enables easy renaming of the output column(s).

FANG_rolling_corr <- returns_joined %>%
    tq_transmute_xy(x          = monthly.returns.x, 
                    y          = monthly.returns.y,
                    mutate_fun = runCor,
                    n          = 6,
                    col_rename = "rolling.corr.6")

And, we can plot the rolling correlations for the FANG stocks.

FANG_rolling_corr %>%
    ggplot(aes(x = date, y = rolling.corr.6, color = symbol)) +
    geom_hline(yintercept = 0, color = palette_light()[[1]]) +
    geom_line(size = 1) +
    labs(title = "FANG: Six Month Rolling Correlation to XLK",
         x = "", y = "Correlation", color = "") +
    facet_wrap(~ symbol, ncol = 2) +
    theme_tq() + 
    scale_color_tq()

Example 4: Use TTR MACD to Visualize Moving Average Convergence Divergence

In reviewing the available options in the TTR package, we see that MACD will get us the Moving Average Convergence Divergence (MACD). In researching the documentation, the return is in the same periodicity as the input and the functions work with OHLC functions, so we can use tq_mutate(). MACD requires a price, so we select close.

FANG_macd <- FANG %>%
    group_by(symbol) %>%
    tq_mutate(select     = close, 
              mutate_fun = MACD, 
              nFast      = 12, 
              nSlow      = 26, 
              nSig       = 9, 
              maType     = SMA) %>%
    mutate(diff = macd - signal) %>%
    select(-(open:volume))
FANG_macd

And, we can visualize the data like so.

FANG_macd %>%
    filter(date >= as_date("2016-10-01")) %>%
    ggplot(aes(x = date)) + 
    geom_hline(yintercept = 0, color = palette_light()[[1]]) +
    geom_line(aes(y = macd, col = symbol)) +
    geom_line(aes(y = signal), color = "blue", linetype = 2) +
    geom_bar(aes(y = diff), stat = "identity", color = palette_light()[[1]]) +
    facet_wrap(~ symbol, ncol = 2, scale = "free_y") +
    labs(title = "FANG: Moving Average Convergence Divergence",
         y = "MACD", x = "", color = "") +
    theme_tq() +
    scale_color_tq()

Example 5: Use xts apply.quarterly to Get the Max and Min Price for Each Quarter

The xts::apply.quarterly() function that is part of the period apply group can be used to apply functions by quarterly time segments. Because we are seeking a return structure that is on a different time scale than the input (quarterly versus daily), we need to use a transmute function. We select tq_transmute and pass the close price using select, and we send this subset of the data to the apply.quarterly function via the mutate_fun argument. Looking at the documentation for apply.quarterly, we see that we can pass a function to the argument, FUN. We want the maximum values, so we set FUN = max. The result is the quarters returned as a date and the maximum closing price during the quarter returned as a double.

FANG_max_by_qtr <- FANG %>%
    group_by(symbol) %>%
    tq_transmute(select     = adjusted, 
                 mutate_fun = apply.quarterly, 
                 FUN        = max, 
                 col_rename = "max.close") %>%
    mutate(year.qtr = paste0(year(date), "-Q", quarter(date))) %>%
    select(-date)
FANG_max_by_qtr

The minimum each quarter can be retrieved in much the same way. The data frames can be joined using left_join to get the max and min by quarter.

FANG_min_by_qtr <- FANG %>%
    group_by(symbol) %>%
    tq_transmute(select     = adjusted, 
                 mutate_fun = apply.quarterly, 
                 FUN        = min, 
                 col_rename = "min.close") %>%
    mutate(year.qtr = paste0(year(date), "-Q", quarter(date))) %>%
    select(-date)

FANG_by_qtr <- left_join(FANG_max_by_qtr, FANG_min_by_qtr,
                         by = c("symbol"   = "symbol",
                                "year.qtr" = "year.qtr"))
FANG_by_qtr

And, we can visualize the data like so.

FANG_by_qtr %>%
    ggplot(aes(x = year.qtr, color = symbol)) +
    geom_segment(aes(xend = year.qtr, y = min.close, yend = max.close),
                 size = 1) +
    geom_point(aes(y = max.close), size = 2) +
    geom_point(aes(y = min.close), size = 2) +
    facet_wrap(~ symbol, ncol = 2, scale = "free_y") +
    labs(title = "FANG: Min/Max Price By Quarter",
         y = "Stock Price", color = "") +
    theme_tq() +
    scale_color_tq() +
    scale_y_continuous(labels = scales::dollar) +
    theme(axis.text.x = element_text(angle = 90, hjust = 1),
          axis.title.x = element_blank())

Example 6: Use zoo rollapply to visualize a rolling regression

A good way to analyze relationships over time is using rolling calculations that compare two assets. Pairs trading is a common mechanism for similar assets. While we will not go into a pairs trade analysis, we will analyze the relationship between two similar assets as a precursor to a pairs trade. In this example we will analyze two similar assets, Mastercard (MA) and Visa (V) to show the relationship via regression.

Before we analyze a rolling regression, it's helpful to view the overall trend in returns. To do this, we use tq_get() to get stock prices for the assets and tq_transmute() to transform the daily prices to daily returns. We'll collect the data and visualize via a scatter plot.

# Get stock pairs
stock_prices <- c("MA", "V") %>%
    tq_get(get  = "stock.prices",
           from = "2015-01-01",
           to   = "2016-12-31") %>%
    group_by(symbol) 

stock_pairs <- stock_prices %>%
    tq_transmute(select     = adjusted,
                 mutate_fun = periodReturn,
                 period     = "daily",
                 type       = "log",
                 col_rename = "returns") %>%
    spread(key = symbol, value = returns)

We can visualize the relationship between the returns of the stock pairs like so.

stock_pairs %>%
    ggplot(aes(x = V, y = MA)) +
    geom_point(color = palette_light()[[1]], alpha = 0.5) +
    geom_smooth(method = "lm") +
    labs(title = "Visualizing Returns Relationship of Stock Pairs") +
    theme_tq()

We can get statistcs on the relationship from the lm function. The model is highly correlated with a p-value of essential zero. The coefficient estimate for V (Coefficient 1) is 0.8134 indicating a positive relationship, meaning as V increases MA also tends to increase.

lm(MA ~ V, data = stock_pairs) %>%
    summary()

While this characterizes the overall relationship, it's missing the time aspect. Fortunately, we can use the rollapply function from the zoo package to plot a rolling regression, showing how the model coefficent varies on a rolling basis over time. We calculate rolling regressions with tq_mutate() in two additional steps:

  1. Create a custom function
  2. Apply the function with tq_mutate(mutate_fun = rollapply)

First, create a custom regression function. An important point is that the "data" will be passed to the regression function as an xts object. The timetk::tk_tbl function takes care of converting to a data frame.

regr_fun <- function(data) {
    coef(lm(MA ~ V, data = timetk::tk_tbl(data, silent = TRUE)))
}

Now we can use tq_mutate() to apply the custom regression function over a rolling window using rollapply from the zoo package. Internally, the returns_combined data frame is being passed automatically to the data argument of the rollapply function. All you need to specify is the mutate_fun = rollapply and any additional arguments necessary to apply the rollapply function. We'll specify a 90 day window via width = 90. The FUN argument is our custom regression function, regr_fun. It's extremely important to specify by.column = FALSE, which tells rollapply to perform the computation using the data as a whole rather than apply the function to each column independently. The col_rename argument is used to rename the added columns.

stock_pairs <- stock_pairs %>%
         tq_mutate(mutate_fun = rollapply,
                   width      = 90,
                   FUN        = regr_fun,
                   by.column  = FALSE,
                   col_rename = c("coef.0", "coef.1"))
stock_pairs

Finally, we can visualize the first coefficient like so. A horizontal line is added using the full data set model. This gives us insight as to points in time where the relationship deviates significantly from the long run trend which can be explored for potential pair trade opportunities.

stock_pairs %>%
    ggplot(aes(x = date, y = coef.1)) +
    geom_line(size = 1, color = palette_light()[[1]]) +
    geom_hline(yintercept = 0.8134, size = 1, color = palette_light()[[2]]) +
    labs(title = "MA ~ V: Visualizing Rolling Regression Coefficient", x = "") +
    theme_tq()

Stock returns during this time period.

stock_prices %>%
    tq_transmute(adjusted, 
                 periodReturn, 
                 period = "daily", 
                 type = "log", 
                 col_rename = "returns") %>%
    mutate(wealth.index = 100 * cumprod(1 + returns)) %>%
    ggplot(aes(x = date, y = wealth.index, color = symbol)) +
    geom_line(size = 1) +
    labs(title = "MA and V: Stock Prices") +
    theme_tq() + 
    scale_color_tq()

Example 7: Use Return.clean and Return.excess to clean and calculate excess returns

In this example we use several of the PerformanceAnalytics functions to clean and format returns. The example uses three progressive applications of tq_transmute to apply various quant functions to the grouped stock prices from the FANG data set. First, we calculate daily returns using quantmod::periodReturn. Next, we use Return.clean to clean outliers from the return data. The alpha parameter is the percentage of oultiers to be cleaned. Finally, the excess returns are calculated using a risk-free rate of 3% (divided by 252 for 252 trade days in one year).

FANG %>%
    group_by(symbol) %>%
    tq_transmute(adjusted, periodReturn, period = "daily") %>%
    tq_transmute(daily.returns, Return.clean, alpha = 0.05) %>%
    tq_transmute(daily.returns, Return.excess, Rf = 0.03 / 252)


Try the tidyquant package in your browser

Any scripts or data that you put into this service are public.

tidyquant documentation built on Nov. 16, 2022, 5:08 p.m.