Indicators

Above/below the zero line and the color change (rising/falling bar) indicate the acceleration direction; unlike AO, it measures whether momentum is accelerating or decelerating.

Computed from the chart OHLCV; takes no arguments. accdist is CUMULATIVE (a line) while iii/wvad are per-bar values - do not confuse them.

Takes no parameters; it uses the bars' H/L/C/V values directly. On bars where (H-L)=0, that bar's contribution is taken as zero.

Uses high/low/close. It measures the strength of the trend, not its direction; above 25 is considered a strong trend, and a separate filter is needed for direction.

SMMA = rma. Shifts: jaw 8, teeth 5, lips 3 bars forward. Due to the shift, the first valid values start on the 20th/12th/7th bar respectively.

The result is None for the first length-1 bars; if there is a None value in the window, that bar returns None.

The calculation is done over the (high+low)/2 average price; positive values show upward momentum and negative values downward momentum.

The EMA is computed with the SMA-seeded Wilder convention; the first valid value appears on the bar where the slow period completes.

The window covers the last length+1 bars including this bar. up=100 indicates that a new highest peak occurred on this bar, down=100 indicates that a new lowest bottom occurred on this bar. The first length bars are None.

Uses Wilder's K/R/SI formula; requires O/H/L/C. The first bar returns None because there is no prev close.

It uses high, low and close; it does not take a separate source argument. Commonly used to set stop distances.

A common expectation is that price recovers as it approaches the lower band and pulls back as it approaches the upper band.

The middle band is found with the SMA and the upper/lower bands with ±stddev standard deviation; the result is the (upper-lower)/middle ratio. It shares the same math as Bollinger Bands (bb).

There is no smoothing; you can smooth the output with sma/ema to reduce noise. On bars containing missing (None) values, the result is None.

The ±100 bands are common thresholds; above +100 signals strong upward pressure and below −100 strong downward pressure. Uses H/L/C.

The first slow-1 bars return None due to insufficient data. chaikinosc_series computes ADL internally.

The first valid value appears (p-1)+(q-1) bars later (in practice p+q-1 bars, since the ATR helper requires p bars). It returns None on insufficient bars.

High values (typically >61.8) indicate a ranging/consolidating market, low values (typically <38.2) indicate a strong trend; it measures only trend strength/absence, not direction.

Positive = buying pressure (close in the upper half of the bar + volume), negative = selling. Uses H/L/C/V.

The first length bars return None; if all movement within the window is zero, it returns 0.

No value is produced on bars where the window sum is zero and on the first length-1 bars (None).

A rise above zero is interpreted as a long-term buy signal.

A value near 1 means moving together and a value near −1 means moving inversely; used in pair selection.

Unlike correlation, covariance is UNSCALED (depends on the series units) and is not normalized to -1..1. Use correlation for a comparable measure.

It is the average of three components: the RSI of close, the RSI of the consecutive up/down streak series, and the percent rank of the 1-bar change. The first valid value is produced after the rankLen bar.

The time axis is fixed at 0..length-1 within the window; if the denominator is zero (constant price), it returns 0.0.

Each bar is the maximum of all values so far; the value only rises when a new peak appears and never falls. Gap (na) values are skipped.

Each bar is the minimum of all values so far; the value only falls when a new bottom appears and never rises. Gap (na) values are skipped.

Supports None-safe chaining: warmup Nones in the source series are preserved, and computation is done over the valid tail then mapped back to the original indices.

decycler = source - highpass(source, length); it preserves the low-frequency trend and removes cycles shorter than length.

Because two nested EMAs are used, roughly the first 2*length-1 bars return None.

basis = sma(source, length), dev = mean(|source - basis|) within the window.

Oscillates around the zero line; extreme positive/negative values can point to overbought/oversold zones. The first (length−1) bars are empty due to warmup.

Direction: if +DI > −DI the uptrend dominates. Strength: adx > 20–25 means a meaningful trend. Fields are accessed with a dot.

Common in breakout strategies: enter when price passes .upper, exit when it breaks .lower.

The shift is length/2+1 bars; the first length-1 bars return None. It does not follow the trend, but emphasizes short-term cycle peaks/troughs.

The EMA is computed over the close price; the first (length-1) bars are empty due to insufficient data.

Catches price reversals earlier than sma; in return it may produce more misleading signals during sudden moves.

Positive values indicate that price moves upward easily, negative values indicate easy downward movement, and values near zero indicate little price movement despite high volume.

No dependency beyond the math.log standard library. hl2 = (high+low)/2 is normalized over the last `length` bars; the value is clamped to ±0.999, and the fisher and normalized value are smoothed with the 0.5/0.66/0.67 coefficients (Ehlers convention).

Above the zero line indicates buying pressure, below it indicates selling pressure; the distance measures the strength of the move.

The first length-1 bars are empty due to warmup (None). It is plotted together with the zero line.

Uses a 5-bar window; the first and last two bars cannot be computed (None). The comparison is a strict inequality (equality does not count as a fractal).

Uses the chart's high/low/close series. Based on the fractal dimension, it reacts quickly in a trend and slowly in a range.

Derived from the Alligator; computed with an SMMA (Wilder RMA) over the median price (H+L)/2 and shifted into the future. The value is empty on the first bars (equal to the shift + the period).

Gives a smooth curve with low lag; shows reversals earlier than classic moving averages.

It is computed over closing prices; the sample (N-1) standard deviation of the ln(close/close[1]) log returns is annualized with sqrt(365) and multiplied by 100. The first 'length' bars (while there are not enough returns) produce no value.

Computed from the chart OHLCV; takes no arguments. accdist is CUMULATIVE (a line) while iii/wvad are per-bar values - do not confuse them.

RVI is computed with the swma method; when the denominator is zero, that bar returns None. Inertia > 0 indicates a bullish, < 0 a bearish bias.

Efficiency ratio = |change over length bars| / |sum of the bar-by-bar changes|. A high ER means a trend, and KAMA approaches the fast coefficient; a low ER means a ranging market, and it approaches the slow coefficient.

Middle = ema(close, length); band = middle ± mult·atr(length). Similar to Bollinger but the band is scaled by ATR.

The middle line is ema(source, length) and the band distance is mult*atr(length); the result is 2*mult*atr/middle. A smaller value means a narrowing channel (squeeze), a larger one a widening channel (volatility). It uses high/low/close (for atr).

The J line can exceed the 0..100 range (intentional; an early overbought/oversold signal). If the Highest-Lowest range is zero, that bar is None.

KST above/below the zero line indicates the long-term momentum direction, while KST crossing the signal indicates reversal signals.

Volume force is computed from the trend determined by the direction of the intraday (H+L+C) value and the ratio of the high-low range to the cumulative measure. The first value appears on the slow bar.

Used to track the direction and slope of the trend in a smooth way.

Fields are accessed with a dot. First assign it to a variable: m = macd(close), then use m.macd and m.signal.

Computed over the source (H+L)/2 (hl2). mama crossing fama upward is interpreted as a buy signal, crossing downward as a sell signal.

The inner EMAs have a fixed 9-period length (classic definition). A rise above 27 followed by a return below 26.5 (reversal bulge) signals a possible trend reversal.

k=0.6 is fixed. The first bar is seeded with the raw source, so there is no insufficient-data period (it produces no None).

The first length-1 bars return None due to insufficient data. Unlike the average, the median is robust to outliers (extreme values).

It is like an rsi that also takes volume into account; below 20 and above 80 are the extreme zones.

It looks at the frequency of values in the last length bars; if several values share the top frequency it returns the smallest. The result is empty (na) on the first length-1 bars.

The result is empty (None) on the first length bars because there is not enough data.

A cumulative/multiplicative index seeded at 1.0, so values stay around 1.0 (e.g. 0.78-1.25). Do NOT confuse it with the classic 1000-based variant - this matches referans dil ta.nvi/ta.pvi. NVI is read as "smart money", PVI as "the crowd".

It adds volume when the close rises and subtracts it when it falls; shows the direction of volume flow.

This is the short-name form of the exact same computation as percentile_linear_interpolation: the window is sorted and the percentile is interpolated between two neighboring values. The result is empty on the first length-1 bars.

This is the short-name form of the exact same computation as percentile_nearest_rank: the window is sorted and the actual value at the matching rank is returned (no interpolation). The result is empty on the first length-1 bars.

The window is sorted and the percentile is found by a weighted transition (interpolation) between two neighboring values; the result need not equal an actual value in the window. pct 50 gives the median. The result is empty on the first length-1 bars.

The window is sorted and the actual value at the rank matching the percentile is returned; no interpolation is done, so the result is always one of the window's observations. The result is empty on the first length-1 bars.

The ratio of values within the last length bars that are less than or equal to the current value is computed as a percentage.

Since atr_series contains a None-warmup, it is not passed directly to ema; the valid tail is taken, ema is computed, and mapped back to the original indices. Because close is raw OHLC, no extra precaution is needed for sma. If the denominator is zero, that bar is None.

It is aligned to the intraday bar; used as a support/resistance reference.

Levels are computed from the high/low/close of the last COMPLETED period (the bars between the two starts marked by anchor). If type is omitted "Traditional" is used. Woodie and DM also account for the open; DM produces only Pivot, R1 and S1, with the rest empty.

A pivot must be strictly greater than all of the leftbars bars to its left and the rightbars bars to its right (equality does not count as a pivot). Because right-side confirmation is required, the value appears rightbars bars after the pivot bar.

The center bar must make a lower low than all of the leftbars bars to its left and the rightbars bars to its right. The value is written to the bar where the pivot is confirmed (center + rightbars); therefore there is no look-ahead. It is the low counterpart of the Pivot High indicator.

Same logic as MACD but uses a percentage relative to the slow EMA instead of an absolute difference; this brings symbols at different price levels to a comparable scale.

For accessing the previous day's close on an intraday chart.

A cumulative/multiplicative index seeded at 1.0, so values stay around 1.0 (e.g. 0.78-1.25). Do NOT confuse it with the classic 1000-based variant - this matches referans dil ta.nvi/ta.pvi. NVI is read as "smart money", PVI as "the crowd".

Same calculation logic as PPO, applied to volume instead of close. A positive PVO indicates that the fast volume average is above the slow one (increasing volume interest).

Standard cumulative PVT math is used. The first bar starts from 0; if the prev close is 0, that bar's contribution is skipped to avoid division by zero. Similar to OBV, but it weights by the percentage price change rather than the full volume.

ema_series starts with an SMA (Wilder), so the first valid value is on the length-1 bar.

A positive value indicates buyer dominance, a negative value indicates seller dominance; the first length-1 bars are None.

It is computed as highest(source, length) − lowest(source, length); a volatility/range measure in price units. The result is empty on the first length-1 bars.

A value near +100 indicates a strong uptrend, a value near -100 indicates a strong downtrend. For equal prices the average rank is used.

The stdev of close is population based (divided by N). The up and down standard deviations are smoothed separately with Wilder RMA; therefore the first stdevLen-1 + length bars return None.

The smoothing used inside rsi and atr. At the same period it reacts more slowly than ema.

The first valid value appears on the momentum + length - 1 bar (bar 17 with the defaults).

Measures momentum; a positive value shows the speed of a rise and a negative value the speed of a fall.

It is a two-stage band-pass filter: first a 2-pole high-pass with period hp filters the trend/DC component, then a Super Smoother with period lp filters the high-frequency noise. It is recursive; on a constant input the output falls to zero.

Above 70 is considered overbought and below 30 oversold. It is more reliable evaluated together with the trend direction rather than on its own.

SWMA has a fixed length of 4; it returns None on the first bars and when the denominator is zero.

The first length bars return None. atr(k) is computed with Wilder ATR; the existing atr_series helper is called directly.

The first bar is None; the trend starts from the second bar. AF increases by inc at each new extreme point, is capped by max, and resets to the start value on a trend reversal.

The result is empty (na) on the first length-1 bars because there is not enough data.

Uses the chart's H/L/C series. It returns None for the first length-1 bars plus the double-EMA lag. On bars where den=0, smi is None.

The first bars are empty until the slow EMA is ready (warmup equal to the slow length). The values come from MACD's double-stage stochastic cycle; they saturate at 0 or 100 in strong trends.

Commonly used for the Z-score: (close − sma(close,n)) / stdev(close,n).

Below 20 is considered oversold and above 80 overbought. Uses high/low/close.

First RSI is computed, then a stochastic is applied to the RSI; identical to stoch(stochLen, smoothK, smoothD, rsi, rsi, rsi). It is more sensitive than plain RSI because it measures RSI's position within its own range.

A 2-pole Butterworth Super Smoother; the coefficient sum is 1 (DC gain 1), and it converges to the same constant on a constant input.

When .direction changes sign, the trend reverses. .line sits below price in an uptrend and above price in a downtrend.

The length is fixed (4) and the weights are [1,2,2,1]/6; therefore it does not take a length parameter. The first 3 bars return None due to warmup.

Because of the chain of six consecutive EMAs, roughly the first 6*(length-1) bars produce no value.

TEMA compensates in a compound manner for the lag created by applying the EMA three times; it reacts faster to price reversals than the classic EMA.

The 50 reference line separates direction: when the value is above 50, upward pressure dominates; when below, downward pressure dominates. If all deviations are zero, the result is taken as 50.

max(H−L, |H−prev C|, |L−prev C|); no smoothing. It is the unsmoothed form of ATR.

Computed with two nested SMAs: first an SMA with a ceil((n+1)/2) window, then an SMA with a floor(n/2)+1 window. The total warmup time is length-1 bars.

Due to the triple EMA, roughly the first 3*(length-1)+1 bars return None; division by zero is guarded.

Identical to linreg(source, length, offset=-1); offset=-1 projects the line onto the next bar (x=length).

Momentum is the difference of the source from the previous bar; because of the double EMA smoothing, roughly the first long+short+1 bars return None.

.on is None for the first length+1 bars due to ATR warmup; .mom returns None for the first ~2*length-2 bars, since linreg warmup is added on top of the source warmup. The BB standard deviation is population (biased) based, consistent with the bbands convention.

Penalizes only declines (drawdowns); unlike standard deviation, upward movement does not increase risk. A higher value = a deeper/longer decline.

Buying pressure BP = close - min(low, prev close); true range TR = max(high, prev close) - min(low, prev close). For each period the average = Σ(BP)/Σ(TR); UO = 100*(4*fast + 2*mid + slow)/7. Standard Williams math is applied.

When the condition is true the source value is captured and held on later bars; n=0 gives the most recent occurrence, n=1 the previous one. If the condition has never occurred yet the result is empty (na).

Variance is the average of the squared deviations from the mean (moving average); it equals the square of the standard deviation (stdev) and measures the dispersion of the series.

The first length-1 bars are None; if the window's sum of absolute changes is zero, that bar is None.

The smoothing coefficient k = (2/(length+1)) * |CMO|; CMO approaches 0 in a ranging market and locks onto the average source.

The source volume is the chart's volume; it does not take a separate source. Positive = volume is above the slow average.

Crossovers of VI+ and VI- are read as trend-reversal signals; both typically oscillate around 1.0.

A reference level commonly used in intraday trading; whether price is above or below the VWAP gives a directional cue.

Weight = bar volume; low-volume bars affect the average less.

The first length-1 bars return None due to warmup. If the EMA volume is zero, that bar is None (division-by-zero protection).

Does not use volume; it computes cumulative accumulation/distribution using only H/L/C. The first bar starts at 0 as the cumulative base.

Crossovers of wt1 and wt2 are used as signals; the extreme zones indicate overbought/oversold

Uses the chart's H/L/C data; the first (length-1) bars return None due to insufficient data.

Above −20 is interpreted as overbought and below −80 as oversold.

The most recent bar is given the highest weight; weights increase from 1 up to length.

It uses high/low/close. The scale is inverted: values near 0 mark the top (overbought) and values near −100 the bottom (oversold); above −20 is overbought and below −80 oversold.

Computed from the chart OHLCV; takes no arguments. accdist is CUMULATIVE (a line) while iii/wvad are per-bar values - do not confuse them.

lag = floor((length-1)/2). A lag-compensation term is added to the source (src + (src - src[lag])), then the EMA is applied.

The standard deviation is population (biased) based. If the standard deviation in the window is zero, the value returns 0.0.