In this vignette I demonstrate how to identify Japanese candlestick patterns on cryptocurrency OHLC data using Bitcoin (BTC) from the {cryptoQuotes}-package.
str(talib::BTC)
#> 'data.frame': 366 obs. of 5 variables:
#> $ open : num 42274 44185 44966 42863 44191 ...
#> $ high : num 44200 45918 45521 44799 44392 ...
#> $ low : num 42181 44152 40555 42651 42362 ...
#> $ close : num 44185 44966 42863 44191 44179 ...
#> $ volume: num 831 2076 2225 1791 2483 ...I begin with a short primer on candlestick construction. I then
define two key control parameters for pattern
recognition—lookbacks (how many prior bars a rule may
inspect) and sensitivity (how strict the rule is with
respect to body/shadow proportions). I finish with charting the
patterns, and give short intepretation of it.
A primer on Candlestick Patterns
A candlestick shows four prices for a chosen time period: Open, High, Low, and Close (OHLC).
The real body is the segment between the Open and the Close. If the Close is above the Open, the candle is bullish; if the Close is below the Open, the candle is bearish. Charting platforms typically color bullish candles green or hollow and bearish candles red or filled, but the color scheme is platform specific.
The line from the high down to the top of the real body is the upper shadow; the line from the low up to the bottom of the real body is the lower shadow. Together they show the price extremes outside the Open-Close range.
Candlestick patterns are built from one or more candles by comparing the size and position of bodies and shadows relative to each other and to the recent trend.
Candlestick patterns in {talib}
In this section I will focus on three patterns: doji(),
harami() and three_outside(). I will start
identifying them with the default lookback and
sensitivity, and then modify the parameters individually to
demonstrate how it affects the number of identified patterns.
## construct data.frame
## with default lookback
## and sensitivity
x <- cbind(
talib::doji(talib::BTC),
talib::harami(talib::BTC),
talib::three_outside(talib::BTC)
)Each function returns a matrix of <integer>: -1
for an identified bearish pattern, 1 for an identified (possibly)
bullish pattern and 0 for no identified pattern. In this vignette we
focus on identified patterns, and therefore only summarize the absolute
values.
NOTE: Some patterns are neither inherently bullish or bearish, and such patterns will always be 1 for identified patterns. A negative value is always bearish.
The identified patterns can be summarized as follows:
{
cat("With default options\n")
apply(abs(x), 2, sum, na.rm = TRUE)
}
#> With default options
#> CDLDOJI CDLHARAMI CDL3OUTSIDE
#> 56 19 21The default settings identified 62, 21 and 21 Doji, Harami and Three
Outside patterns. There are an important aspect to take note of here:
<NA>-values.
The <NA>-values are a byproduct of the underlying
algorithms in identifying the patterns. The doji(), for
example, uses a default lookback period of 10 candles in identifying the
pattern—and therefore discards the first 10 values. The same logic
applies for harami() and three_outside().
Candle inclusion (lookback)
The default number of candles used to identify a specific can be
modified with options(). The doji() and
harami() depends on the candle bodies which can be modified
as follows:
## set included number
## of candles
options(
talib.BodyDoji.N = 5,
talib.BodyLong.N = 7,
talib.BodyShort.N = 7
)To determine whether the real body is that of a Doji, it will now use the past 5 candles, instead of the default 10 candles. And similarly it will use the past 7 candles to determine if the real body is long or short. The identified patterns can be summarized as follows:
{
cat("With modified lookback\n")
apply(abs(x), 2, sum, na.rm = TRUE)
}
#> With modified lookback
#> CDLDOJI CDLHARAMI CDL3OUTSIDE
#> 62 21 21By modifying the settings 62, 21 and 21 Doji, Harami and Three Outside patterns. Notice that the Three Outside pattern did not change; this is because the pattern always needs three candles to be identified.
Candle sensitivity
Before proceeding to analyze candle sensitivity, I set the candlestick lookbacks to its default values:
## default values
options(
talib.BodyDoji.N = 10,
talib.BodyLong.N = 10,
talib.BodyShort.N = 10
)Candle sensitivity in {talib} is a sensitivity factor that multiplies the reference value from the previous N candles (lookback); the current candle is tested against . A smaller makes the condition harder to satisfy.
The sensitivity can be modified as follows:
## set sensitivity
## of candles
options(
talib.BodyDoji.alpha = 0.2
)The identified patterns can be summarized as follows:
{
cat("With modified sensitivity options\n")
apply(abs(x), 2, sum, na.rm = TRUE)
}
#> With modified sensitivity options
#> CDLDOJI CDLHARAMI CDL3OUTSIDE
#> 109 19 21The default setting for doji() is 0.1, and setting this
value higher increases the identified number of Doji-patterns to 109. In
this case the 0.2 value translates to the following:
Real body is like Doji’s body when it’s shorter than 20% the average of the 10 previous candles’ high-low range.
Charting
The candlestick patterns can be charted similar to the remaining indicators in {talib}.
{
## candlestick chart
talib::chart(talib::BTC)
## chart identified 'doji'-patterns
talib::indicator(talib::doji)
## chart identified 'harami'-patterns
talib::indicator(talib::harami)
## chart identified 'three outside'-patterns
talib::indicator(talib::three_outside)
}Identified patterns that are neither bullish or bearish are colored in a neutral way, while bullish or bearish patterns follows the typical color conventions.
Contributing and bug reports
The underlying library, TA-Lib, is a long-standing and well-known library but this R wrapper is still in its early stage. All contributions, suggestions, and bug reports are welcome.