Williams AO/AC Signal Methodology

The Williams AO/AC Indicator: Technical Specification

The Williams Radar uses two oscillators developed by Bill Williams, described in Trading Chaos (1995) and New Trading Dimensions (1998). This page documents the exact formulas, signal classification, scanning logic, and the backtested base rates we measured before going live.

What Is the Awesome Oscillator (AO)?

The Awesome Oscillator measures the difference between short-term and long-term price momentum using the midpoint of each bar.

Formula:

Midpoint = (High + Low) / 2
AO = SMA(5, Midpoint) - SMA(34, Midpoint)

Where SMA(n, x) is the simple moving average of x over the last n periods.

Interpretation:

• AO > 0: short-term momentum stronger than long-term trend — bullish bias.
• AO < 0: short-term momentum weaker than long-term — bearish bias.
• AO crossing zero from below: potential momentum shift.

What Is the Accelerator Oscillator (AC)?

The Accelerator Oscillator measures the rate of change of market driving force. It is derived from the AO.

Formula:

AC = AO - SMA(5, AO)

The AC leads the AO — it changes direction before the AO does.

Interpretation:

• AC turning positive while AO rising: acceleration phase — strongest entry zone.
• AC turning negative while AO still positive: deceleration — momentum stalling.
• Both AO and AC positive and aligned: full confirmation signal.

Signal Classification

Signal classification is not a buy/sell recommendation. See the Disclaimer.

Scanning Universe

255 US-listed equities across 13 sector ETFs covering the major segments of the US market: XLU, XLI, XLY, XLV, XLRE, XLK, XLF, XLC, XLB, IBB, XLP, XBI, XLE. The full ETF-to-sector mapping is on the About page.

The scanner runs every Friday after US market close (18:00 Mexico City time). It fetches weekly OHLCV data from Alpha Vantage, calculates AO and AC per ticker, and classifies matching signals.

Data source: Alpha Vantage — adjusted weekly closes.

Universe Selection Criteria

Tickers are not selected by individual screening — they enter the universe by belonging to a sector ETF, then pass two additional filters:

What is explicitly NOT used as a selection filter:

• Dividend yield — high-yield tickers are not preferred over low-yield ones
• Price-to-earnings or valuation multiples — no fundamental screening
• Analyst ratings or price targets
• Recent momentum or trend — tickers are not added or removed based on recent performance

The universe is additive — tickers are added weekly from the expansion schedule (rank 21+ per sector, in order of sector quality), and discarded only when an S1 signal is ≥ 20 weeks old with no AO confirmation, or following an extraordinary structural break (e.g., bankruptcy filing).

Ranging Filter

Before a ticker is evaluated for S1/S2 signals, it passes through a ranging filter:

• Calculated as: price range over the last 12 weeks < 15% of the 12-week average price
• If a ticker is lateralizing (low volatility, compressed range), it is excluded from all signal categories — S1, S2 Pure, and S2 Degraded
• Rationale: AO/AC signals in ranging markets produce false positives at a much higher rate than in trending markets

Two additional price context fields are computed but do not exclude a ticker — they inform ranking only:

• nearLows: whether the current price is within 15% of its 104-week low (2-year range)
• pricePercentile: where the current price sits within its 104-week high/low range (0% = at lows, 100% = at highs)

Weekly Output

1. CSV file — signal table: Ticker, Sector, AO, AC, Signal type, Week label.
2. Journal entry — published at thewilliamsradar.com/w{NN}-{YYYY} with narrative analysis.
3. Telegram notification — delivered immediately after scan completes.

Backtested Base Rates (2001–2026)

Before publishing the radar live, we ran the same AO/AC logic against 25 years of weekly data — 2001 through 2026 — to measure base rates with the same rules we use now. The numbers below are the results of those tests; they are why we publish, and they are what the live record will be measured against.

Sector ETF backtest

338 signals across 8 sector ETFs, 2001–2026 weekly data:

Individual ticker backtest

3,774 S1 signals across 79 individual tickers (XLU, XLP, XLE, XLI constituents), 2001–2026:

S2 confirmation tightens the cohort

491 S2 signals (−87 % vs S1's 3,774) across the same 79-ticker universe. S2 requires two consecutive confirmation bars rather than one, materially shrinking the count of fired entries and concentrating on the higher-quality configurations.

Key insight from the backtest: in individual tickers, AO confirmation lags S1 by an average of 10–18 weeks. The patient investor has 2–4 months of observation before the AO confirms the move — patience is part of the system, not a flaw of it.

Known Limitations of the Backtest

Two limitations we are explicit about:

1. Survivorship bias — the universe used for the historical backtest is the current (2026) membership of XLU / XLI / XLP / XLE. Historical results from 2001–2026 therefore ignore every name that was delisted, went bankrupt, or dropped out of its sector ETF in that window. Future versions will reconstruct point-in-time membership.
2. Lookahead bias (now fixed) — the original backtests filled the trade at the close of the signal bar — the same bar whose AO/AC was used to decide the signal. The current engine fills at the next bar's open, which is what a live trader can actually do.

The live record published from W17 onward is run under the same engine that produced the corrected backtests above — no retrofit, no edits.

What This Is Not

The Williams Radar does not:

• Predict price targets or stop levels.
• Account for position sizing or portfolio construction.
• Incorporate fundamental analysis, earnings, or macro factors.
• Constitute investment advice of any kind.

For questions, reach out via X/Twitter @fedemoctezuma.