@pond-ts/financial

Technical studies and a trading calendar on pond-ts.

Twenty oracle-verified studies (moving averages, bands, RSI, MACD, ATR, stochastics, Donchian, OBV, VWAP, …) that append columns to a bar TimeSeries, plus a TradingCalendar that knows when the market is open so rolling windows, bucketing and chart axes stop at the close. Pure computation: browser + Node, no data fetching, no rendering, no React (the chart side lives in @pond-ts/charts).

npm install @pond-ts/financial pond-ts

pond-ts is a peer dependency; the pond packages release together, so keep their ranges in step.

Quick start

A study takes a series and options and returns the series with more columns on it. Import the fluent entry once and chain them:

import '@pond-ts/financial/fluent';

const study = bars // a TimeSeries with open/high/low/close/volume columns
  .bollinger({ period: 20 })
  .ema({ period: 10 })
  .rsi({ period: 14 })
  .macd({ fastPeriod: 12, slowPeriod: 26, signalPeriod: 9 })
  .vwap({ period: 20 });
// + bbUpper/bbMiddle/bbLower, ema, rsi, macdLine/macdSignal/macdHist, vwap

Every study is also a plain function, for code that would rather not augment TimeSeries:

import { bollinger, ema } from '@pond-ts/financial';

const same = ema(bollinger(bars, { period: 20 }), { period: 10 });

Each study reads a column (default 'close', or the named high / low / close / volume inputs for the multi-input ones) so it runs over any numeric column, including another study's output. Periods are bar counts. Warm-up rows are undefined and the row count is preserved, so the result lines up on the source's time axis.

A trading calendar turns "5-minute bars" into session-aligned bars — no weekend or holiday buckets, no bar spanning the close:

import { TradingCalendar } from '@pond-ts/financial';

const cal = TradingCalendar.fromRules(
  { timeZone: 'America/New_York', open: '09:30', close: '16:00' },
  { from: '2026-01-05', to: '2026-02-13' },
);

const fiveMin = ticks.aggregate(cal.barSequence('5m'), {
  close: { from: 'price', using: 'last' },
});
cal.isOpen(instant); // inside a session and not inside a break

What's in the box

• Studies — verified bar-for-bar against a pandas oracle before they ship, the named indicators against TA-Lib as well:
  • averages and bands: sma, ema, bollinger, envelope, donchian, vwap
  • oscillators: rsi, macd, stochastic, williamsR, momentum, percentChange, zScore, obv
  • volatility and range: atr, historicalVolatility, rollingStdev, rollingMin, rollingMax, rollingPercentile
• @pond-ts/financial/fluent — mounts every study as a TimeSeries method (opt-in by import; ESM only).
• TradingCalendar — fromRules (hours, weekmask, holidays, breaks, earlyCloses, resolved DST-correctly in the exchange's time zone) or fromSessions (an explicit list); sessions(), sessionOn(), isTradingDay(), isOpen(), sessionSequence() / barSequence() for bucketing, and tagSessions() to key a partitionBy so stateful ops never bridge a close.
• Discontinuity providers — weekendSkip(), segmentDiscontinuity(), identityDiscontinuity() and calendar.discontinuities(): the d3fc-style five-method surface a trading-time axis consumes. @pond-ts/charts reads it structurally, so there is no package coupling.
• @pond-ts/financial/parallel — Node-only: .withWorkers() runs the rolling studies across worker threads for large partitioned series.

Bar-for-bar vendor parity is a non-goal: where a study deliberately departs from TA-Lib (a flat window is undefined, not 0; a gap in a running sum propagates rather than being skipped) the docstring says so and a test pins it.

Documentation

Guides, live examples and the full API live at https://pond-ts.org — the package page and the financial charts hub. Source and issues: github.com/pond-ts/pond.

License

MIT