B3 is the main financial exchange in Brazil, offering support and access to trading systems for equity and fixed income markets. In its website you can find a vast number of datasets regarding prices and transactions for contracts available for trading at these markets, including:
- equities (unadjusted for corporate events)
- futures
- FII (Reits)
- options
- BDRs
- historical yield curves (calculated from futures contracts)
- B3 indexes composition
Package rb3 facilitates downloading and reading these datasets from B3, making it easy to consume it in R in a structured way. These datasets can be used in industry or academic studies.
The documentation is available in its pkgdown page, where articles (vignettes) with real applications can be found.
install.packages("rb3")
# github (Development branch)
if (!require(devtools)) install.packages("devtools")
devtools::install_github("wilsonfreitas/rb3")
Download and use historical yield curve data with yc_get
.
library(rb3)
library(ggplot2)
library(stringr)
df_yc <- yc_mget(
first_date = Sys.Date() - 255 * 5,
last_date = Sys.Date(),
by = 255
)
p <- ggplot(
df_yc,
aes(
x = forward_date,
y = r_252,
group = refdate,
color = factor(refdate)
)
) +
geom_line() +
labs(
title = "Yield Curves for Brazil",
subtitle = "Built using interest rates future contracts",
caption = str_glue("Data imported using rb3 at {Sys.Date()}"),
x = "Forward Date",
y = "Annual Interest Rate",
color = "Reference Date"
) +
theme_light() +
scale_y_continuous(labels = scales::percent)
print(p)
Get settlement future prices with futures_get
.
library(rb3)
library(dplyr)
df <- futures_mget(
first_date = "2022-04-01",
last_date = "2022-04-29",
by = 5
)
glimpse(
df |>
filter(commodity == "DI1")
)
#> Rows: 153
#> Columns: 8
#> $ refdate <date> 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04-01, 2022-04~
#> $ commodity <chr> "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1", "DI1"~
#> $ maturity_code <chr> "J22", "K22", "M22", "N22", "Q22", "U22", "V22", "X22", "Z22", "F23", "G23", "H23", "J23", "N23", "V23", "F24", "J24", "N24", "V24", "F25", "J25", "N25"~
#> $ symbol <chr> "DI1J22", "DI1K22", "DI1M22", "DI1N22", "DI1Q22", "DI1U22", "DI1V22", "DI1X22", "DI1Z22", "DI1F23", "DI1G23", "DI1H23", "DI1J23", "DI1N23", "DI1V23", "D~
#> $ price_previous <dbl> 99999.99, 99172.50, 98159.27, 97181.87, 96199.14, 95137.64, 94174.49, 93265.23, 92365.48, 91404.64, 90434.90, 89662.14, 88719.94, 86306.50, 84065.98, 82~
#> $ price <dbl> 100000.00, 99172.31, 98160.23, 97185.43, 96210.42, 95159.25, 94209.42, 93314.08, 92422.80, 91472.04, 90513.73, 89751.01, 88821.78, 86457.18, 84241.81, 8~
#> $ change <dbl> 0.01, -0.19, 0.96, 3.56, 11.28, 21.61, 34.93, 48.85, 57.32, 67.40, 78.83, 88.87, 101.84, 150.68, 175.83, 275.55, 370.18, 423.00, 452.44, 477.65, 497.68,~
#> $ settlement_value <dbl> 0.01, 0.19, 0.96, 3.56, 11.28, 21.61, 34.93, 48.85, 57.32, 67.40, 78.83, 88.87, 101.84, 150.68, 175.83, 275.55, 370.18, 423.00, 452.44, 477.65, 497.68, ~
Equity closing data (without ANY price adjustments) is available
thru cotahist_get
.
library(rb3)
library(bizdays)
# fix for ssl error (only in linux)
if (Sys.info()["sysname"] == "Linux") {
httr::set_config(
httr::config(ssl_verifypeer = FALSE)
)
}
date <- preceding(Sys.Date() - 1, "Brazil/ANBIMA") # last business day
ch <- cotahist_get(date, "daily")
glimpse(
cotahist_equity_get(ch)
)
#> Rows: 381
#> Columns: 13
#> $ refdate <date> 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 20~
#> $ symbol <chr> "AALR3", "ABCB4", "ABEV3", "AERI3", "AESB3", "AFLT3", "AGRO3", "AGXY3", "ALLD3", "ALPA3", "ALPA4", "ALPK3", "ALSO3", "ALUP3", "ALUP4", "ALUP11", "A~
#> $ open <dbl> 19.77, 16.81, 14.72, 3.89, 10.81, 9.16, 31.62, 9.94, 12.25, 18.35, 21.29, 3.83, 20.11, 8.82, 8.82, 26.40, 2.40, 30.30, 22.94, 5.71, 34.83, 12.98, 8~
#> $ high <dbl> 20.06, 16.93, 14.82, 3.95, 10.89, 9.16, 31.99, 10.30, 12.43, 18.60, 21.63, 3.83, 20.21, 8.92, 8.88, 26.60, 2.48, 31.78, 23.50, 5.76, 36.00, 13.15, ~
#> $ low <dbl> 19.74, 16.41, 14.18, 3.76, 10.66, 9.16, 30.45, 9.94, 11.66, 17.90, 20.74, 3.67, 19.16, 8.73, 8.76, 26.22, 2.31, 30.06, 22.56, 5.45, 33.79, 12.72, 8~
#> $ close <dbl> 20.06, 16.61, 14.22, 3.78, 10.68, 9.16, 30.62, 9.98, 11.81, 17.91, 20.94, 3.67, 19.35, 8.88, 8.76, 26.40, 2.35, 31.19, 22.95, 5.54, 35.33, 13.00, 8~
#> $ average <dbl> 19.89, 16.59, 14.42, 3.82, 10.74, 9.16, 30.99, 10.07, 11.89, 18.19, 21.04, 3.78, 19.45, 8.83, 8.80, 26.41, 2.38, 31.04, 22.97, 5.57, 35.67, 12.95, ~
#> $ best_bid <dbl> 20.05, 16.60, 14.22, 3.77, 10.68, 9.06, 30.62, 9.98, 11.81, 17.90, 20.91, 3.66, 19.35, 8.87, 8.75, 26.34, 2.35, 31.04, 22.92, 5.53, 34.00, 12.98, 8~
#> $ best_ask <dbl> 20.09, 16.61, 14.23, 3.78, 10.71, 9.46, 30.70, 10.00, 11.98, 20.00, 20.95, 3.82, 19.40, 8.88, 8.80, 26.40, 2.36, 31.19, 22.95, 5.54, 35.98, 13.03, ~
#> $ volume <dbl> 10070975, 11310732, 425796280, 7981542, 9264003, 6412, 17228579, 502600, 1295768, 136426, 80283647, 212998, 44464947, 76829, 61642, 41557139, 14061~
#> $ traded_contracts <int> 506100, 681700, 29516600, 2087900, 862300, 700, 555900, 49900, 108900, 7500, 3814800, 56300, 2286000, 8700, 7000, 1573300, 5884000, 1319900, 102964~
#> $ transactions_quantity <int> 2069, 4754, 50558, 4973, 3369, 3, 2778, 312, 623, 25, 19031, 149, 9519, 64, 45, 6068, 4665, 6909, 21590, 3080, 1180, 3316, 5567, 13942, 111, 5901, ~
#> $ distribution_id <int> 102, 140, 125, 101, 102, 119, 112, 101, 103, 231, 231, 101, 102, 113, 113, 113, 113, 102, 101, 107, 105, 103, 142, 104, 105, 101, 101, 119, 119, 10~
Funds data
glimpse(
cotahist_etfs_get(ch)
)
#> Rows: 81
#> Columns: 13
#> $ refdate <date> 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 20~
#> $ symbol <chr> "5GTK11", "ACWI11", "ALUG11", "ASIA11", "BBOV11", "BBSD11", "BDIV11", "BITH11", "BOVA11", "BOVB11", "BOVS11", "BOVV11", "BOVX11", "BRAX11", "BRZP11~
#> $ open <dbl> 80.00, 9.11, 37.50, 7.30, 56.40, 91.65, 92.00, 35.60, 104.78, 108.49, 83.27, 109.48, 10.89, 90.51, 65.57, 53.80, 11.08, 8.84, 25.19, 71.25, 33.00, ~
#> $ high <dbl> 80.00, 9.11, 37.50, 7.37, 56.40, 91.65, 92.90, 36.50, 104.78, 108.49, 83.27, 109.48, 10.89, 90.51, 66.67, 53.80, 11.08, 8.84, 25.39, 71.75, 33.05, ~
#> $ low <dbl> 77.10, 8.86, 36.85, 7.22, 54.45, 89.60, 91.96, 34.18, 102.15, 106.54, 81.54, 106.81, 10.61, 89.21, 65.15, 53.80, 10.75, 8.39, 23.00, 70.25, 31.96, ~
#> $ close <dbl> 78.25, 8.90, 36.95, 7.27, 54.45, 89.60, 92.55, 35.40, 102.21, 106.70, 81.71, 106.86, 10.63, 89.30, 65.16, 53.80, 10.77, 8.50, 23.20, 70.25, 32.30, ~
#> $ average <dbl> 78.85, 8.90, 37.15, 7.26, 54.45, 89.85, 92.20, 34.91, 102.94, 106.78, 82.18, 108.19, 10.66, 89.89, 65.35, 53.80, 10.79, 8.66, 23.48, 70.94, 32.49, ~
#> $ best_bid <dbl> 77.11, 8.85, 36.85, 7.22, 54.15, 89.50, 92.00, 35.00, 102.21, 106.70, 81.71, 106.86, 10.63, 89.30, 65.16, 0.00, 10.76, 8.30, 23.06, 70.25, 32.30, 8~
#> $ best_ask <dbl> 78.25, 8.90, 36.95, 7.28, 55.57, 91.25, 92.88, 35.40, 102.26, 108.49, 0.00, 107.64, 11.08, 90.50, 65.77, 53.80, 11.00, 8.50, 24.50, 71.09, 38.00, 1~
#> $ volume <dbl> 4.770952e+04, 3.425857e+06, 1.001285e+05, 1.671910e+06, 2.237427e+07, 3.037244e+04, 7.305553e+05, 8.654818e+05, 1.053651e+09, 1.836637e+06, 4.15840~
#> $ traded_contracts <int> 605, 384777, 2695, 230241, 410880, 338, 7923, 24791, 10235510, 17200, 506, 2480688, 389143, 966, 4926, 1, 10161, 8548, 6455, 29741, 1167, 6, 108, 6~
#> $ transactions_quantity <int> 8, 463, 64, 57, 40, 21, 712, 908, 71821, 65, 140, 39872, 568, 16, 1244, 1, 42, 105, 130, 1562, 10, 2, 2, 18, 31, 8, 6, 7, 16, 3392, 159, 51, 26, 9,~
#> $ distribution_id <int> 100, 100, 100, 100, 100, 100, 107, 100, 106, 100, 100, 100, 100, 100, 127, 100, 100, 100, 100, 100, 100, 100, 100, 100, 102, 100, 100, 100, 100, 10~
FIIs data
glimpse(
cotahist_fiis_get(ch)
)
#> Rows: 263
#> Columns: 13
#> $ refdate <date> 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 20~
#> $ symbol <chr> "BZLI11", "ABCP11", "AFHI11", "AFOF11", "AIEC11", "ALMI11", "ALZR11", "APTO11", "ARCT11", "ARRI11", "ATSA11", "BARI11", "BBFI11B", "BBFO11", "BBPO1~
#> $ open <dbl> 16.00, 72.75, 98.37, 89.10, 79.78, 950.00, 116.38, 10.19, 105.17, 94.28, 90.20, 99.79, 1967.00, 73.40, 81.50, 98.00, 64.89, 81.95, 106.70, 98.98, 9~
#> $ high <dbl> 16.90, 72.77, 99.31, 91.49, 79.78, 950.00, 116.40, 10.19, 105.17, 95.05, 90.20, 99.99, 1967.00, 73.40, 81.80, 98.00, 65.00, 82.99, 107.34, 98.98, 9~
#> $ low <dbl> 16.00, 72.26, 98.08, 88.51, 78.83, 938.60, 115.99, 9.80, 104.72, 94.00, 90.20, 98.23, 1927.01, 71.50, 81.50, 97.00, 64.40, 80.81, 106.55, 98.00, 9.~
#> $ close <dbl> 16.90, 72.76, 99.01, 88.95, 79.64, 950.00, 116.35, 9.99, 105.14, 94.03, 90.20, 99.00, 1939.95, 71.50, 81.76, 97.67, 64.91, 81.13, 107.00, 98.00, 9.~
#> $ average <dbl> 16.67, 72.67, 98.49, 89.55, 79.39, 947.15, 116.25, 9.92, 105.01, 94.31, 90.20, 99.21, 1941.05, 72.49, 81.67, 97.68, 64.81, 81.59, 106.79, 98.65, 9.~
#> $ best_bid <dbl> 16.90, 72.68, 98.69, 88.95, 79.60, 938.60, 116.34, 9.99, 105.13, 94.06, 88.01, 98.98, 1937.77, 71.23, 81.51, 97.31, 64.91, 81.13, 107.00, 85.01, 9.~
#> $ best_ask <dbl> 17.17, 72.76, 99.01, 89.00, 79.64, 954.00, 116.35, 10.11, 105.14, 94.46, 89.12, 99.00, 1939.95, 71.50, 81.76, 97.67, 64.92, 81.22, 107.25, 98.00, 9~
#> $ volume <dbl> 66.70, 47457.74, 763930.66, 245015.64, 411115.49, 7577.20, 771576.07, 14696.45, 2275005.35, 258529.97, 90.20, 1249031.26, 296981.79, 146510.85, 100~
#> $ traded_contracts <int> 4, 653, 7756, 2736, 5178, 8, 6637, 1481, 21664, 2741, 1, 12589, 153, 2021, 12276, 580, 29721, 8715, 8294, 3, 9484, 3, 155, 13163, 56974, 56, 1434, ~
#> $ transactions_quantity <int> 2, 113, 882, 274, 702, 8, 1034, 116, 2654, 200, 1, 1667, 56, 122, 803, 64, 5372, 1349, 806, 2, 495, 3, 14, 1446, 2417, 7, 23, 55, 1325, 13, 6146, 2~
#> $ distribution_id <int> 100, 314, 114, 114, 120, 250, 154, 105, 127, 131, 136, 137, 302, 116, 214, 228, 252, 185, 190, 129, 106, 123, 115, 120, 116, 102, 225, 216, 211, 11~
BDRs data
glimpse(
cotahist_bdrs_get(ch)
)
#> Rows: 504
#> Columns: 13
#> $ refdate <date> 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 20~
#> $ symbol <chr> "A1AP34", "A1BB34", "A1CR34", "A1DI34", "A1DM34", "A1EE34", "A1EN34", "A1EP34", "A1KA34", "A1LG34", "A1LL34", "A1LN34", "A1MB34", "A1MD34", "A1MP34~
#> $ open <dbl> 60.47, 36.09, 65.42, 409.20, 428.56, 232.99, 299.40, 250.50, 40.29, 338.30, 61.75, 32.09, 381.37, 499.99, 322.60, 571.52, 50.21, 126.30, 200.97, 29~
#> $ high <dbl> 60.47, 36.09, 65.52, 410.80, 428.56, 233.54, 301.38, 252.50, 40.29, 338.30, 61.75, 32.09, 381.37, 515.00, 322.60, 581.09, 50.21, 126.30, 200.97, 29~
#> $ low <dbl> 60.47, 36.09, 63.63, 399.59, 426.34, 231.14, 295.49, 249.00, 40.11, 338.30, 60.90, 32.09, 381.37, 478.42, 322.60, 554.38, 50.21, 126.30, 198.24, 29~
#> $ close <dbl> 60.47, 36.09, 63.63, 400.40, 426.34, 231.42, 295.49, 250.49, 40.11, 338.30, 60.90, 32.09, 381.37, 479.57, 322.60, 554.40, 50.21, 126.30, 199.92, 29~
#> $ average <dbl> 60.47, 36.09, 64.47, 404.83, 427.46, 232.21, 298.17, 251.23, 40.20, 338.30, 61.20, 32.09, 381.37, 483.72, 322.60, 561.21, 50.21, 126.30, 199.53, 29~
#> $ best_bid <dbl> 0.00, 0.00, 54.00, 0.00, 400.00, 0.00, 0.00, 0.00, 34.75, 312.00, 60.29, 0.00, 0.00, 449.99, 0.00, 534.37, 0.00, 0.00, 0.00, 275.80, 195.02, 28.50,~
#> $ best_ask <dbl> 0.00, 38.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 925.00, 64.01, 42.48, 0.00, 479.56, 0.00, 0.00, 52.08, 200.00, 0.00, 0.00, 250.00, 49.05, 0.~
#> $ volume <dbl> 60.47, 72.18, 20762.45, 42507.94, 255198.22, 16255.32, 16399.72, 55523.72, 160.80, 4736.20, 2080.80, 641.80, 381.37, 920537.51, 6452.00, 175097.52,~
#> $ traded_contracts <int> 1, 2, 322, 105, 597, 70, 55, 221, 4, 14, 34, 20, 1, 1903, 20, 312, 1, 34, 6, 578, 29, 15, 58, 86, 5, 306, 560, 13723, 112, 1, 1, 5, 673, 369974, 12~
#> $ transactions_quantity <int> 1, 1, 206, 101, 9, 70, 55, 70, 2, 2, 4, 1, 1, 189, 1, 104, 1, 1, 6, 151, 5, 1, 58, 4, 1, 224, 217, 28, 1, 1, 1, 2, 6, 3550, 105, 132, 3, 101, 206, ~
#> $ distribution_id <int> 110, 102, 106, 109, 110, 109, 110, 110, 101, 100, 111, 100, 110, 100, 110, 109, 103, 101, 110, 109, 109, 100, 109, 103, 100, 109, 103, 100, 101, 10~
Equity options
glimpse(
cotahist_equity_options_get(ch)
)
#> Rows: 6,711
#> Columns: 14
#> $ refdate <date> 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 20~
#> $ symbol <chr> "ABCBE175", "ABCBQ160", "ABEVB15", "ABEVD30", "ABEVE121", "ABEVE125", "ABEVE127", "ABEVE130", "ABEVE132", "ABEVE135", "ABEVE137", "ABEVE140", "ABEV~
#> $ type <fct> Call, Put, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call, Call~
#> $ strike <dbl> 17.23, 15.73, 14.47, 29.47, 11.04, 12.54, 12.79, 13.04, 13.29, 13.54, 13.79, 14.04, 14.29, 14.54, 14.79, 15.04, 15.29, 15.54, 15.79, 16.04, 16.29, ~
#> $ maturity_date <date> 2022-05-20, 2022-05-20, 2023-02-17, 2023-04-20, 2022-05-20, 2022-05-20, 2022-05-20, 2022-05-20, 2022-05-20, 2022-05-20, 2022-05-20, 2022-05-20, 20~
#> $ open <dbl> 0.01, 0.14, 2.34, 0.01, 3.19, 2.20, 1.85, 1.52, 1.47, 1.08, 0.83, 0.77, 0.43, 0.40, 0.13, 0.06, 0.02, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.0~
#> $ high <dbl> 0.01, 0.14, 2.34, 0.01, 3.20, 2.20, 1.85, 1.65, 1.47, 1.15, 0.83, 0.77, 0.43, 0.43, 0.15, 0.06, 0.02, 0.02, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.0~
#> $ low <dbl> 0.01, 0.14, 2.34, 0.01, 3.19, 2.20, 1.48, 1.25, 1.00, 0.73, 0.46, 0.27, 0.14, 0.05, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.0~
#> $ close <dbl> 0.01, 0.14, 2.34, 0.01, 3.20, 2.20, 1.51, 1.25, 1.00, 0.73, 0.46, 0.28, 0.15, 0.06, 0.02, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.0~
#> $ average <dbl> 0.01, 0.14, 2.34, 0.01, 3.19, 2.20, 1.51, 1.46, 1.11, 0.95, 0.68, 0.51, 0.24, 0.14, 0.04, 0.02, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.01, 0.0~
#> $ volume <dbl> 1, 14, 234, 3, 959, 1100, 83275, 139792, 145053, 213675, 24886, 38081, 102317, 134668, 42983, 20926, 7175, 370, 227, 574, 11, 42, 5, 1000, 52, 2240~
#> $ traded_contracts <int> 100, 100, 100, 300, 300, 500, 55000, 95300, 130400, 223700, 36300, 74100, 413700, 935300, 1006500, 730300, 565800, 36900, 22700, 57400, 1100, 4200,~
#> $ transactions_quantity <int> 1, 1, 1, 1, 2, 1, 11, 23, 10, 265, 19, 69, 79, 194, 116, 238, 77, 35, 110, 15, 11, 42, 3, 13, 52, 1, 4, 16, 1, 25, 12, 159, 29, 300, 65, 202, 227, ~
#> $ distribution_id <int> 139, 139, 124, 124, 122, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 125, 124, 12~
The list with available B3 indexes can be obtained with indexes_get
.
indexes_get()
#> [1] "AGFS" "BDRX" "GPTW" "IBOV" "IBRA" "IBXL" "IBXX" "ICO2" "ICON" "IDIV" "IEEX" "IFIL" "IFIX" "IFNC" "IGCT" "IGCX" "IGNM" "IMAT" "IMOB" "INDX" "ISEE" "ITAG" "IVBX" "MLCX" "SMLL"
#> [26] "UTIL"
And the composition of a specific index with index_comp_get
.
(ibov_comp <- index_comp_get("IBOV"))
#> [1] "ABEV3" "ALPA4" "AMER3" "ASAI3" "AZUL4" "B3SA3" "BBAS3" "BBDC3" "BBDC4" "BBSE3" "BEEF3" "BIDI11" "BPAC11" "BPAN4" "BRAP4" "BRFS3" "BRKM5" "BRML3" "CASH3"
#> [20] "CCRO3" "CIEL3" "CMIG4" "CMIN3" "COGN3" "CPFE3" "CPLE6" "CRFB3" "CSAN3" "CSNA3" "CVCB3" "CYRE3" "DXCO3" "ECOR3" "EGIE3" "ELET3" "ELET6" "EMBR3" "ENBR3"
#> [39] "ENEV3" "ENGI11" "EQTL3" "EZTC3" "FLRY3" "GGBR4" "GOAU4" "GOLL4" "HAPV3" "HYPE3" "IGTI11" "IRBR3" "ITSA4" "ITUB4" "JBSS3" "JHSF3" "KLBN11" "LCAM3" "LREN3"
#> [58] "LWSA3" "MGLU3" "MRFG3" "MRVE3" "MULT3" "NTCO3" "PCAR3" "PETR3" "PETR4" "PETZ3" "POSI3" "PRIO3" "QUAL3" "RADL3" "RAIL3" "RDOR3" "RENT3" "RRRP3" "SANB11"
#> [77] "SBSP3" "SLCE3" "SOMA3" "SULA11" "SUZB3" "TAEE11" "TIMS3" "TOTS3" "UGPA3" "USIM5" "VALE3" "VBBR3" "VIIA3" "VIVT3" "WEGE3" "YDUQ3"
With the index composition you can use COTAHIST to select their quotes.
glimpse(
cotahist_get_symbols(ch, ibov_comp)
)
#> Rows: 92
#> Columns: 13
#> $ refdate <date> 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 2022-05-18, 20~
#> $ symbol <chr> "ABEV3", "ALPA4", "AMER3", "ASAI3", "AZUL4", "B3SA3", "BBAS3", "BBDC3", "BBDC4", "BBSE3", "BEEF3", "BIDI11", "BPAC11", "BPAN4", "BRAP4", "BRFS3", "~
#> $ open <dbl> 14.72, 21.29, 22.94, 15.81, 21.79, 11.72, 36.40, 16.26, 19.83, 25.79, 12.46, 15.37, 24.83, 8.59, 26.94, 14.23, 44.09, 9.05, 2.01, 13.10, 3.35, 11.0~
#> $ high <dbl> 14.82, 21.63, 23.50, 15.83, 22.20, 11.82, 36.88, 16.37, 19.95, 25.88, 12.46, 15.37, 24.83, 8.71, 27.06, 14.64, 44.09, 9.09, 2.08, 13.19, 3.46, 11.2~
#> $ low <dbl> 14.18, 20.74, 22.56, 15.38, 21.06, 11.47, 35.73, 15.93, 19.47, 25.27, 11.94, 14.24, 23.61, 8.46, 26.12, 13.58, 42.00, 8.63, 1.94, 12.81, 3.31, 10.9~
#> $ close <dbl> 14.22, 20.94, 22.95, 15.40, 21.23, 11.55, 35.91, 16.01, 19.56, 25.41, 12.19, 14.32, 23.87, 8.50, 26.22, 13.58, 42.59, 8.69, 1.95, 13.02, 3.38, 11.0~
#> $ average <dbl> 14.42, 21.04, 22.97, 15.52, 21.61, 11.61, 36.20, 16.06, 19.60, 25.54, 12.19, 14.55, 23.94, 8.52, 26.42, 14.02, 42.81, 8.75, 2.00, 12.98, 3.38, 11.1~
#> $ best_bid <dbl> 14.22, 20.91, 22.92, 15.39, 21.23, 11.52, 35.91, 16.00, 19.56, 25.40, 12.19, 14.31, 23.85, 8.50, 26.20, 13.58, 42.51, 8.69, 1.94, 12.97, 3.38, 11.0~
#> $ best_ask <dbl> 14.23, 20.95, 22.95, 15.40, 21.25, 11.56, 35.92, 16.01, 19.57, 25.41, 12.20, 14.32, 23.87, 8.51, 26.22, 13.65, 42.60, 8.71, 1.95, 13.02, 3.39, 11.0~
#> $ volume <dbl> 425796280, 80283647, 236561952, 106264437, 151706028, 531192301, 452153919, 184075368, 893663060, 111365341, 106738532, 280265348, 348285985, 61860~
#> $ traded_contracts <int> 29516600, 3814800, 10296400, 6842800, 7019100, 45745200, 12487200, 11457500, 45581800, 4359900, 8754000, 19251700, 14547800, 7257400, 3577300, 1184~
#> $ transactions_quantity <int> 50558, 19031, 21590, 13942, 16706, 48810, 34065, 19677, 47357, 15329, 14078, 33793, 28525, 11778, 12391, 22330, 11918, 44802, 11339, 27195, 17874, ~
#> $ distribution_id <int> 125, 231, 101, 104, 101, 121, 305, 741, 741, 120, 113, 110, 112, 113, 143, 120, 120, 117, 102, 141, 149, 231, 104, 102, 134, 143, 111, 123, 256, 11~
One important part of rb3
infrastructure is its Template System
.
All datasets handled by the rb3 package are configured in a template, that is an YAML file. The template brings many information regarding the datasets, like its description and its metadata that describes its columns, their types and how it has to be parsed. The template fully describes its dataset.
Once you have the template implemented you can fetch and read downloaded
data directly with the functions download_marketdata
and
read_marketdata
.
For examples, let’s use the template FPR
to download and read data
regarding primitive risk factor used by B3 in its risk engine.
f <- download_marketdata("FPR", refdate = as.Date("2022-05-10"))
f
#> [1] "C:/Users/wilso/AppData/Local/Temp/RtmpKc4zmX/rb3-cache/FPR-7a2422cc97221426a3b2bd4419215481/FP220510/FatoresPrimitivosRisco.txt"
download_marketdata
returns the path for the downloaded file.
fpr <- read_marketdata(f, "FPR")
fpr
#> $Header
#> # A tibble: 1 x 2
#> tipo_registro data_geracao_arquivo
#> <int> <date>
#> 1 1 2022-05-10
#>
#> $Data
#> # A tibble: 3,204 x 11
#> tipo_registro id_fpr nome_fpr formato_variacao id_grupo_fpr id_camara_indicador id_instrumento_indicador origem_instrumento base base_interpolacao criterio_capitalizacao
#> <int> <int> <chr> <fct> <dbl> <chr> <dbl> <int> <int> <int> <int>
#> 1 2 1422 VLRAPT4 Basis Points 1 BVMF 200000008810 8 0 0 0
#> 2 2 1423 VLPETR3 Basis Points 1 BVMF 200000008803 8 0 0 0
#> 3 2 1424 VLSEER3 Basis Points 1 BVMF 200000008818 8 0 0 0
#> 4 2 1426 VLJBSS3 Basis Points 1 BVMF 200000008780 8 0 0 0
#> 5 2 1427 VLKLBN11 Basis Points 1 BVMF 200000008781 8 0 0 0
#> 6 2 1428 VLITUB3 Basis Points 1 BVMF 200000463163 8 0 0 0
#> 7 2 1429 VLITSA4 Basis Points 1 BVMF 200000008777 8 0 0 0
#> 8 2 1430 VLHYPE3 Basis Points 1 BVMF 200000008773 8 0 0 0
#> 9 2 1431 VLGRND3 Basis Points 1 BVMF 200000008770 8 0 0 0
#> 10 2 1433 VLUGPA3 Basis Points 1 BVMF 200000008830 8 0 0 0
#> # ... with 3,194 more rows
#>
#> attr(,"class")
#> [1] "parts"
read_marketdata
parses the downloaded file according to the metadata
configured in the template FPR
.
Here it follows a view of the show_templates
adding that lists the
available templates.
show_templates()