pandas_ta Column Names
Which columns each indicator actually appends to your DataFrame. Captured by running the library against real price data, not copied from documentation.
Why BBM_20_2.0 raises a KeyError today
The 0.3.x release line is no longer installable from PyPI. A fresh `pip install pandas_ta` today gives 0.4.x, whose generated column names differ from every tutorial written against 0.3.x.
| Indicator | Name in 0.3.x tutorials | Measured in 0.4.71b0 |
|---|---|---|
| bbands | BBL_20_2.0 / BBM_20_2.0 / BBU_20_2.0 | BBL_5_2.0_2.0 / BBM_5_2.0_2.0 / BBU_5_2.0_2.0 |
default length changed 20 -> 5, and the suffix gained a fourth component. The 0.3.x names are recorded here as they appear in tutorials — they could not be re-measured, because that release is no longer installable (pip install pandas_ta==0.3.14b0 in a clean venv returned 'No matching distribution found ... (from versions: 0.4.67b0, 0.4.71b0)'). Only the 0.4.71b0 column is measured.
Every column, measured on 0.4.71b0
Defaults only — change a parameter and the suffix changes with it. Run df.ta.bbands(length=20) and the column becomes BBM_20_...; the suffix is the parameters.
| Call | Cols | Columns created |
|---|---|---|
| df.ta.bbands() | 5 | BBL_5_2.0_2.0, BBM_5_2.0_2.0, BBU_5_2.0_2.0, BBB_5_2.0_2.0, BBP_5_2.0_2.0 |
| df.ta.macd() | 3 | MACD_12_26_9, MACDh_12_26_9, MACDs_12_26_9 |
| df.ta.stoch() | 3 | STOCHk_14_3_3, STOCHd_14_3_3, STOCHh_14_3_3 |
| df.ta.stochrsi() | 2 | STOCHRSIk_14_14_3_3, STOCHRSId_14_14_3_3 |
| df.ta.adx() | 4 | ADX_14, ADXR_14_2, DMP_14, DMN_14 |
| df.ta.atr() | 1 | ATRr_14 |
| df.ta.rsi() | 1 | RSI_14 |
| df.ta.ema() | 1 | EMA_10 |
| df.ta.sma() | 1 | SMA_10 |
| df.ta.wma() | 1 | WMA_10 |
| df.ta.vwap() | 1 | VWAP_D |
| df.ta.obv() | 1 | OBV |
| df.ta.cci() | 1 | CCI_14_0.015 |
| df.ta.mfi() | 1 | MFI_14 |
| df.ta.willr() | 1 | WILLR_14 |
| df.ta.psar() | 4 | PSARl_0.02_0.2, PSARs_0.02_0.2, PSARaf_0.02_0.2, PSARr_0.02_0.2 |
| df.ta.supertrend() | 4 | SUPERT_7_3.0, SUPERTd_7_3.0, SUPERTl_7_3.0, SUPERTs_7_3.0 |
| df.ta.kc() | 3 | KCLe_20_2, KCBe_20_2, KCUe_20_2 |
| df.ta.donchian() | 3 | DCL_20_20, DCM_20_20, DCU_20_20 |
| df.ta.aroon() | 3 | AROOND_14, AROONU_14, AROONOSC_14 |
| df.ta.ao() | 1 | AO_5_34 |
| df.ta.apo() | 1 | APO_12_26 |
| df.ta.bop() | 1 | BOP |
| df.ta.cmf() | 1 | CMF_20 |
| df.ta.coppock() | 1 | COPC_11_14_10 |
| df.ta.dm() | 2 | DMP_14, DMN_14 |
| df.ta.efi() | 1 | EFI_13 |
| df.ta.eom() | 1 | EOM_14_100000000 |
| df.ta.kdj() | 3 | K_9_3, D_9_3, J_9_3 |
| df.ta.kst() | 2 | KST_10_15_20_30_10_10_10_15, KSTs_9 |
| df.ta.massi() | 1 | MASSI_9_25 |
| df.ta.mom() | 1 | MOM_10 |
| df.ta.natr() | 1 | NATR_14 |
| df.ta.ppo() | 3 | PPO_12_26_9, PPOh_12_26_9, PPOs_12_26_9 |
| df.ta.pvo() | 3 | PVO_12_26_9, PVOh_12_26_9, PVOs_12_26_9 |
| df.ta.qstick() | 1 | QS_10 |
| df.ta.roc() | 1 | ROC_10 |
| df.ta.rvi() | 1 | RVI_14 |
| df.ta.slope() | 1 | SLOPE_1 |
| df.ta.squeeze() | 4 | SQZ_20_2.0_20_1.5, SQZ_ON, SQZ_OFF, SQZ_NO |
| df.ta.trix() | 2 | TRIX_30_9, TRIXs_30_9 |
| df.ta.tsi() | 2 | TSI_13_25_13, TSIs_13_25_13 |
| df.ta.uo() | 1 | UO_7_14_28 |
| df.ta.vortex() | 2 | VTXP_14, VTXM_14 |
| df.ta.zscore() | 1 | ZS_30 |
How to read a suffix
pandas_ta names each output NAME_param1_param2...using the parameters the call actually ran with — which is why the name changes when you change a default, and why hardcoding a column string breaks silently on upgrade. The robust pattern is to take the returned frame's columns rather than typing the name: bb = df.ta.bbands(); mid = bb.iloc[:, 1], or bb.filter(like='BBM').
Measured on 400 real AAPL bars (2025-01-15 to 2026-08-19), pandas_ta 0.4.71b0, Python 3.12.10. Regenerate with python -m pipeline.pandas_ta_columns.