TA-Lib/ta-lib v0.8.1

on GitHub

Installation docs: https://ta-lib.org/install/

Just re-install to upgrade. Older versions are automatically removed.

See github commits for complete list of changes

[0.8.1] 2026-09-12

Added

• New Streaming API. See https://ta-lib.org/api/stream/
• (#81) Microsoft VCPKG support. Thanks @greenTableWork !
• (#78) CMake can now opt out of building the static or the shared library (both built by default). Thanks @BwL1289 !
• (#75) More docs for DEMA, TEMA, T3, MFI, ULTOSC, KAMA and TRIX. Thanks @nehemiah888 !
• New TA Functions:
  • AC: Accelerator/Decelerator Oscillator (#228)
  • ADR: Average Day Range, the mean of the last n bar ranges (#367)
  • AO: Awesome Oscillator (#227)
  • CMF: Chaikin Money Flow (#134)
  • CMOU: Chande Momentum Oscillator, Unsmoothed (#124)
  • CVI: Chaikin's Volatility (#358)
  • COPPOCK: Coppock Curve (#362)
  • CUMSUM: Cumulative Sum (#372)
  • DONCHIAN: Donchian Channels (#342)
  • DPO: Detrended Price Oscillator (#363)
  • ER: Kaufman Efficiency Ratio (#350)
  • ERI: Elder Ray Index, Bull Power / Bear Power (#361)
  • EFI: Elder's Force Index (#206)
  • FOSC: Forecast Oscillator (#345)
  • FRACTAL: Williams Fractal (#371)
  • HA: Heikin-Ashi Candles, an OHLC-to-OHLC smoothing transform (#373)
  • HMA: Hull Moving Average (#139)
  • KC: Keltner Channels (#273)
  • KDJ: KDJ Stochastic (#365)
  • MARKETFI: Market Facilitation Index (#230)
  • MASSI: Mass Index (#359)
  • NVI: Negative Volume Index (#126)
  • PERCENTILE: Percentile, statistic of the trailing window (#368)
  • PERCENTRANK: Percent Rank, share of the previous window a value ranks above (#369)
  • PVI: Positive Volume Index (#126)
  • PVO: Percentage Volume Oscillator (#119)
  • PVT: Price Volume Trend (#364)
  • QSTICK: Qstick, from Tushar Chande and Stanley Kroll (#226)
  • RMA: Wilder's Smoothed Moving Average (#348)
  • RVI: Relative Volatility Index (#366)
  • RVOL: Relative Volume (#370)
  • SMI: Stochastic Momentum Index (#238)
  • SUPERTREND: SuperTrend, an ATR-scaled trailing band with a trend flag (#272)
  • TSI: True Strength Index (#360)
  • VHF: Vertical Horizontal Filter (#346)
  • VORTEX: Vortex Indicator (#349)
  • VWAP: Volume Weighted Average Price (#237)
  • VWMA: Volume Weighted Moving Average (#131)
  • WAD: Williams' Accumulation/Distribution (#200)
  • ZLEMA: Zero-Lag Exponential Moving Average (#347)
• New MAType (for MA, BBANDS, STOCH etc...):
  • TA_MAType_HMA (#139)
  • TA_MAType_DISABLED — no smoothing at any period; output copy the input (#93)
  • TA_MAType_DEFAULT — selects that parameter's documented MA type (#182)
  • TA_MAType_ZLEMA (#347)
  • TA_MAType_RMA (#348)

Faster

• ~8x: MACD, MACDFIX and MACDEXT (when MA type is EMA).
• ~3x to 7x: DEMA, TEMA and TRIX
• ~2x to 4x: ATR and NATR (#338)
• ~2x: RSI (#410)
• ~2x: ACCBANDS, MFI (#244), SQRT (#192). Thanks @kevinlincg !
• ~1.6x to 15x: MIN, MAX, MINMAX, MIDPOINT, MIDPRICE and WILLR (#147). Thanks @kevinlincg !
• ~40%: ULTOSC (#154). Thanks @dexhunter !
• ~30%: MAVP (#143). Thanks @dexhunter !
• ~27% Apple, ~8% GCC: AROON, AROONOSC and others (#128). Thanks @dexhunter !
• ~20%: VAR, STDDEV, BBANDS

Changed

• (#133) BBANDS default optInTimePeriod changed from 5 to 20, as intended by John Bollinger.
• (#120) PPO and APO now default optInMAType to EMA (was SMA), matching Gerald Appel's original PPO/MACD definition. Pass TA_MAType_SMA explicitly to keep the previous behavior.
• (#96) Fused multiply-add and other floating-point re-ordering produce minor output differences; an intentional modernization.
• (#338) ATR, NATR smooth the true range with a fused two-coefficient step, thus changing
  the order of operations. Values moved by at most 1.3e-15 from legacy results.
• (#183) EMA now uses a fused multiply-add in its recursion, as the EMA cascades inside
  DEMA, TEMA, TRIX, MACD and MACDFIX already did. Values move by at most 2.8e-16 relative
  from the reference series, and the same shift reaches MA, BBANDS, APO, PPO, PVO, MAVP,
  STOCH, STOCHF and STOCHRSI when the MAtype is EMA.
• (#410) RSI optimization with some floating-point re-ordering. On the 0-100 scale
  values changed by at most 5.7e-14 at the default period, growing slowly with the period
  (7.8e-14 at 100, 5.3e-13 at 20000) and unaffected by the price scale. STOCHRSI inherits
  the change, measured up to 1.1e-11 different than the legacy calculation.
• (#4,#14) API: TA_FUNC_UNST_MFI and TA_FUNC_UNST_IMI enum constants removed
• (#129) API: TA_FUNC_UNST_ADXR and TA_FUNC_UNST_STOCHRSI enum constants removed.
• (#180) API: startIdx and endIdx are now capped at the new TA_MAX_INDEX (100,000,000);
  above it a call returns TA_OUT_OF_RANGE_START_INDEX / TA_OUT_OF_RANGE_END_INDEX.
• (#144) API: TA_FUNC_UNST_NONE enum constant removed. It could not be passed in
  (it is rejected) and was never returned, so it had no use in the public API.
• (#122) Removed the ide/ directory (Visual Studio/Xcode/MSVC project files). Use autotools, CMake and vcpkg instead.
• (#386) API: the shared library's SONAME is now libta-lib.so.1. The two build systems
  previously disagreed — autotools produced libta-lib.so.0, CMake libta-lib.so.0.8.1 —
  so a binary built against one could not load the other. Both now derive it from a single
  TALIB_LIBRARY_VERSION in configure.ac, and it changes only when the ABI does, not
  every release. Binaries linked against the old names must be relinked; they will fail to
  load rather than silently misreading TA_FuncInfo.
• (#77) CMake shared library now links libm directly, so it declares its own math-library dependency instead of relying on the consuming program to provide it. Thanks @BwL1289 !

Removed

• (#400) API: the shared library now exports only what the installed headers declare — 2,452
  symbols instead of 3,968. The 1,516 removed were internals no header ever mentioned.
• (#386) API: TA_FuncInfo.camelCaseName. Code that reads it no longer compiles. Code that reads
  any field after it — flags, nbInput, nbOptInput, nbOutput, handle — must be rebuilt:
  the struct is 8 bytes smaller and every one of those fields moves.
• (#388) API: the MetaStock variant when the MAType is EMA, DEMA or TEMA is removed. Default behavior is unchanged. TA_SetCompatibility() and TA_GetCompatibility() remain, so existing sources still compile and link, but the setter now does nothing and the getter always answers TA_COMPATIBILITY_DEFAULT. Moving forward TA-Lib no longer supports "Compatibility Settings".

Fixed

• (#385, #390) On divide by zero, KAMA could return -Inf, after which every remaining bar of the call was NaN, and could return values outside the range of the prices it was smoothing. Output is now always a valid number.
• (#130) In-place calls (same buffer as input and output) returned wrong values for STOCH, STOCHF and MAVP. Regular (separate-buffer) calls were always correct.
• (#118,#242) VAR, CORREL, STDDEV and BBANDS more precise and faster.
• (#33) Float overflow in the single-precision (TA_S_*) functions. Thanks @iglesias !
• (#64) Website docs mixing up CDL3LINESTRIKE with CDL3OUTSIDE's description. Thanks @mw66 !
• (#7) CCI returned a spurious value when all prices over the period were identical; Thanks @trufanov-nok for identifying and resolving this!
• (#57) Missing TA_GetVersionString function in Windows DLL. Thanks @Youngv !
• (#98) TRIX and NATR returned wrong values when startIdx > lookback. NATR additionally left output slots unwritten for any bar with a zero close.
• (#98) A non-zero unstable period settings changed IMI's summation window. Fine with default settings.
• (#107) MFI and STOCHRSI could return a wrong value when floating-point rounding left a near-zero result that was then compared exactly against zero. Thanks @Caleblgx, @trufanov-nok and @mrjbq7 !
• (#4,#14) MFI and IMI are no longer flagged as having an unstable period. Thanks @mw66 and @wony-zheng !
• (#99) BBANDS with TA_MAType_MAMA and a period >= 34 returned a misaligned middle band.
• (#102) Fixed ULTOSC and CDL3INSIDE performance (regression only in 0.7.1)
• (#112) IMI returned NaN on an all-flat window (every bar close == open); now returns 50.0.
• (#202) VAR no longer returns a tiny negative variance on a flat stretch, where the calculation cancels to either side of zero; it now returns 0.0 instead.
• (#243) STDDEV and BBANDS returned exactly 0 for a standard deviation that was small but non-zero. In rare cases, was making the bands "collapse" on the middle line.
• (#244) MFI returned 0 instead of the index whenever the window summed to less than 1.0. Also, no longer returns values slightly outside 0-100 (clamps the epsilon errors).
• (#253) Fix many TA_IS_ZERO vs TA_IS_ZERO_SCALED choices. Numerically better for edge cases, like very small inputs (<10e-8) or mostly flat input prices.
• (#390) STOCH and STOCHF returned inf or NaN while reporting success, for prices near the bottom of the double range. A close sitting on the window high now comes out as exactly 100.
• (#395) On extreme inputs, CCI could return +/-Inf, CORREL a NaN or a perfect +/-1 correlation from a window that had none, and WILLR a value outside its documented [-100, 0] range. CCI and CORREL now answer such a window with 0, and WILLR is always in range, with a close on the period low coming out as exactly -100.