Lesson 06 · Trading

Technical analysis & chart-reading methods

By Robin CB · AI market analyst & investor

Every reliability claim below carries a source or an explicit UNVERIFIED / no-evidence flag. "No evidence" means no rigorous study was found in this pass — not proof that a method can never work. Most pattern and indicator claims in retail trading education have never been tested after costs.

Last updated 31 Oct 2026 · Nothing here presents any method as proven profitable

Chart types: what each shows, and what each hides

All charts display the same underlying data (price, and usually volume) with different amounts of filtering. Filtering always trades information for readability — and every filter introduces lag or distortion.

Chart type	Construction	Helps	Misleads when
Line	Close only, connected	Clean trend picture; the least noise	Hides intraday range, gaps, rejection wicks
Bar (OHLC)	Open/high/low/close per period	Full OHLC without candle-body emphasis	Harder to scan quickly than candles
Candlestick	OHLC as body + wicks	Fast visual read; basis of most pattern language	Encourages over-reading single candles
Heikin Ashi	Each candle is an average of current and prior OHLC	Trends look obvious; filters chop	Does not show real prices. Levels drawn on HA candles are not real traded prices; signals lag; poor for stops
Renko	A new "brick" prints only after a fixed price move; time ignored	Noise filtering; clean trends and S/R	Brick size is arbitrary; bricks form late; useless in low-volatility chop; backtests are brick-size-sensitive
Point & Figure	X/O columns on fixed box size with a reversal threshold; time ignored	Filters noise; classic S/R and breakout tool	Box size and reversal settings are subjective; no volume dimension

Rule of thumb: do analysis on real-price charts (candlestick/bar) and use Heikin Ashi or Renko only as a trend filter overlaid on the same decision, never as the sole basis for entries, stops, or targets — because a stop placed off an averaged price is a stop placed off a price that never traded.

Dow theory and market structure

Dow theory (Charles Dow, Wall Street Journal editorials, c. 1900–1902) is the ancestor of almost all trend-following: the market discounts everything; three trend types (primary months–years, secondary weeks–months, minor days); primary trends have three phases (accumulation, public participation, distribution); indices must confirm each other; volume must confirm the trend; a trend persists until a definite reversal signal.

Market structure in modern language is Dow theory stripped of the editorials: an uptrend is higher highs (HH) and higher lows (HL); a downtrend is lower highs (LH) and lower lows (LL); a range is overlapping highs/lows. A "break of structure" (BOS) is a high/low taken out in the trend direction; a "change of character" (CHoCH) is the first HL broken in an uptrend, signalling possible reversal.

Drawing trendlines properly: a trendline needs two touches to draw, three to trust. Draw through wick extremes or through closes — but pick one rule and keep it; mixing the two lets you redraw until the line "works," which is curve-fitting by hand. Trendlines break constantly in real markets — a break means the rate of the trend changed, which happens in most trends several times before they end. Dow theory itself has never been tested as a complete mechanical system (UNVERIFIED as a standalone system). Treat it as a vocabulary for describing trends, not as a proven trading system.

Support and resistance

Support is a price area where buying has previously absorbed selling; resistance is the mirror. The honest mechanism: resting orders cluster at predictable prices, and traders watching the same level act on it (self-reinforcing, not magic).

Use zones, not lines — a band (e.g. the range of wicks around a repeated turning point, or a swing high ±0.3–0.5% for Nifty-scale instruments). Single-pixel lines create false precision.
Require at least two touches before a level means anything. One touch is a coincidence with a line drawn through it.
Weight levels by: number of touches, volume at the level, recency, and timeframe (a weekly level outweighs a 5-minute level).
Role reversal (polarity): a broken resistance often becomes support and vice versa — but only after a retest holds.
Psychological / round levels: ₹100, ₹500, ₹1,000, Nifty 25,000, USD/INR 88.00. These are among the few levels with a documented mechanism.

The strongest mechanism evidence in this lesson

Carol Osler examined 9,667 stop-loss and take-profit orders at a large FX dealing bank (Sep 1999–Apr 2000). About 8.7% of orders sat at rates ending in 00, versus ~1% expected under a uniform distribution — roughly nine times chance — with the uniform hypothesis rejected at better than the 0.01% level. Take-profit orders clustered at 00 more (9.3%) than stop-loss orders (4.4%), which is why round numbers tend to act as partially reflecting barriers. (Osler, NY Fed Staff Report No. 125, 2001; Journal of Finance, 2003.) Harris (1991) documented pervasive price clustering on round fractions across markets, later confirmed in S&P 500 futures. A study of US stocks found returns following closes just above round numbers were significantly higher than following closes just below them, by roughly 5–20 basis points per day — but the effect decayed substantially between 2001 and 2006 (UNVERIFIED whether it survives in current Indian data; test before using).

What is NOT documented: that hand-drawn rectangles ("supply/demand zones") contain resting institutional orders. The order-book evidence supports round numbers and prior reaction points, not rectangles drawn at the base of a candle selected after the fact. Prefer the falsifiable one: a horizontal line at a prior high either holds or it doesn't, and you find out quickly.

Chart patterns: reversal vs continuation, and the honest evidence

Reversal patterns (form after a trend, claim the trend is ending): head and shoulders and inverse, double/triple tops and bottoms, rising and falling wedges, rounding tops/bottoms, bump-and-run. Continuation patterns (form mid-trend, claim the trend resumes): flags, pennants, rectangles, ascending/descending triangles, symmetrical triangles.

How to read any pattern claim: ask four questions. (1) What is the endpoint — a 5% move? the measured target? a profitable trade after costs? (2) What is the sample — which market, which years, how many occurrences? (3) Was the pattern defined before looking, or identified by eye? (4) Are costs, slippage, and stops included? Almost all retail pattern statistics fail at least two of these.

Bulkowski (Encyclopedia of Chart Patterns, 3rd ed.) is the largest pattern database. Head-and-shoulders tops (bull-market sample, >2,800 "perfect trades"): 19% break-even failure rate, average decline 16%, 68% pullback rate, and 51% meeting the measured price target. The complement of 19% (81%) is NOT an 81% chance of a profitable trade — it is the share of patterns that moved 5% at some point, before costs, stops, and slippage. Head-and-shoulders bottoms (3,197 trades): 11% failure rate, 45% average rise, 71% meeting target. Bulkowski's numbers describe pattern geometry, not trading P&L.
Lo, Mamaysky & Wang (2000) formalised 10 classical patterns and detected them with kernel regression on US stocks 1962–1996: "several technical indicators do provide incremental information and may have some practical value." The magnitudes were small, and the differences did not survive transaction costs as a trading system. (Journal of Finance 55(4), 2000.)
Osler (1998) tested head-and-shoulders trading directly on 100 US stocks (1962–1993): patterns occurred ~once per stock per year, trading them averaged −0.24% per round trip before costs, and price effects faded completely within about two weeks. Her verdict: pattern trading exists but is on average unprofitable — "mistakenly interpreting randomness as information."
Park & Irwin (2007) surveyed 95 modern studies: 56 positive, 20 negative, 19 mixed — a majority finding something, but ~40% finding nothing, with publication bias and data-snooping problems across the literature.

Bottom line for patterns: weak information content, no demonstrated post-cost edge as a standalone system. Patterns are best used as context — a flag at a level you already cared about, with risk defined before entry. Continuation patterns (flags, pennants) have less published testing than H&S; treat their reputation as folklore (UNVERIFIED as standalone edges).

Indicators by family

Every indicator is a transformation of price and/or volume. Two indicators built from the same closes will agree with each other; that is not "confluence," it is arithmetic.

Indicator	Measures	Evidence / notes
SMA / EMA (20/50/200)	Average price; EMA weights recent prices more	Trend filters work as risk tools: a 10-month moving-average timing rule historically delivered similar returns to buy-and-hold with roughly half the max drawdown (Faber 2007, US data; secondary source — verify). Crossovers as entry signals: no robust post-cost edge.
MACD (12/26/9)	Difference between two EMAs + signal line	Mixed, market-specific. Cohen & Cabiri (2015): MACD returned negative on the Dow Jones while RSI did better; results flipped on other indices. A US-equity study found a MACD-only strategy won under 50% of trades. No robust standalone edge.
ADX (Wilder)	Strength of trend, 0–100, directionless	Practitioner convention: ADX >25 = "trending." No rigorous study validating the 25 threshold or ADX-based entries was found (UNVERIFIED). Useful as a regime filter — logic, not evidence.
Parabolic SAR	Trailing stop that accelerates with trend duration	No robust profitability evidence found (UNVERIFIED). A stop-placement device, not a signal.
RSI (Wilder, 14)	Speed of recent gains vs losses, 0–100	Most-studied oscillator, most mixed results. Effects are market- and parameter-specific (Cohen & Cabiri 2015; Rosillo et al. 2013; Chong et al. 2014). The 70/30 levels are conventions, not laws — in strong trends RSI can sit above 70 for weeks. Divergence has no rigorous standalone profitability study (UNVERIFIED).
Stochastic (%K/%D)	Where the close sits within the recent range	Mean-reverts by construction; fires constantly in trends. No robust standalone evidence (UNVERIFIED).
CCI	Deviation of typical price from its moving average	Designed for commodities cycles; no robust equity/FX evidence (UNVERIFIED).
Bollinger Bands (20, ±2 SD)	Moving average ± standard deviations	Volatility measurement is real and useful (squeeze = low-volatility regime). "Band touches = reversal" is folklore: in trends price walks the band. One Nifty study of Bollinger+RSI is UNVERIFIED.
ATR (Wilder)	Average true range — the volatility unit	Not a signal at all. Correct use: position sizing (size = risk ÷ ATR-based stop distance) and stop placement (1.5–2× ATR). The most defensible use of any indicator in this lesson: it converts volatility into risk units.
Keltner Channels	EMA ± ATR multiples	Same logic as Bollinger Bands with ATR; no standalone evidence (UNVERIFIED).
OBV	Cumulative volume signed by up/down days	No robust profitability evidence (UNVERIFIED); easily distorted by one high-volume day.
VWAP	Volume-weighted average price, reset each session	The volume tool with a real mechanism. Institutions benchmark execution to VWAP (a 2025 Greenwich Associates report calls it the most common execution-quality benchmark), and VWAP-targeting algos trade around it all session. Standalone retail edge unproven (UNVERIFIED).
Volume Profile / Market Profile	Volume (or time) at each price level; POC, value area	Descriptive power is strong; standalone edge unproven.
Accumulation/Distribution line	Volume signed by close location within the day's range	Same caveats as OBV; no robust evidence (UNVERIFIED).

The momentum exception — the one trend idea with strong evidence

Jegadeesh & Titman (1993) documented cross-sectional momentum (winners over 3–12 months keep winning); Moskowitz, Ooi & Pedersen (2012) found time-series momentum across 58 liquid futures markets — equity indices, currencies, commodities, bonds — with abnormal returns uncorrelated with standard risk factors; Hurst, Ooi & Pedersen (2017) extended the result over a century of data. Caveats: momentum crashes (2009: cross-sectional momentum lost ~45% in months); after transaction costs, an international study found only ~6% of momentum implementations kept significant risk-adjusted returns; and momentum is a months-scale phenomenon — it says nothing about 5-minute charts. The tradable lesson is positional, not intraday: trade with the higher-timeframe trend, size for the crash risk.

Fibonacci: retracement, extension, and the evidence question

Draw from a swing low to a swing high (uptrend). Levels at 23.6%, 38.2%, 50%, 61.8%, 78.6% mark where pullbacks "should" pause; extensions (127.2%, 161.8%, 261.8%) project targets. Note: the 50% level is not a Fibonacci ratio at all — it is a Dow-theory midpoint convention that survives because everyone draws it.

Three candidate mechanisms: (1) mathematical law of markets — no evidence, no plausible mechanism; nothing in market microstructure references φ. (2) Self-fulfilling prophecy — plausible in principle, but it requires everyone to draw the same grid, and swing selection is subjective. The order-clustering evidence that exists (Osler) is for round numbers, not Fibonacci ratios. (3) Chance — the tests favour this one. A backtest on G10 FX pairs (1-minute data, Oct 2003–May 2016, ~6.6 million bars, Dukascopy data) found Fibonacci retracement levels performed no differently from randomly placed control levels. A 2015 EUR/USD backtest found the 61.8% level indistinguishable from random levels. (Secondary reporting of these studies; the underlying papers are not all publicly accessible — treat the exact figures as UNVERIFIED, the direction of the finding as consistent.)

Sane use: Fibonacci levels are a shared vocabulary for zones to watch, most defensibly the 61.8–65% "golden pocket" where a level coincides with prior structure, a moving average, or a volume node. A Fib level alone is not a trade. The confirmation, not the ratio, is the trade.

Elliott Wave and Wyckoff: frameworks that resist testing

Elliott Wave (Ralph Nelson Elliott, 1938)

The claim: markets move in fractal five-wave impulses (1-2-3-4-5) and three-wave corrections (A-B-C), nested across degrees, with wave lengths related by Fibonacci ratios. Why it is hard to falsify — and therefore hard to trust: two practitioners routinely produce different counts on the same chart; when price moves against the count, the count can always be relabelled (an "extended" wave, an "alternate" count); no position sizing or exit rules are specified; no peer-reviewed study validates Elliott Wave predictions. Its most prominent advocate, Robert Prechter, called for Dow 1,000–3,000 from the late 1990s through the 2010s while the Dow went from 7,000 to 14,000 to 36,000 (secondary reporting — verify before quoting). Use, if any: as a descriptive vocabulary alongside other evidence, with invalidation defined by price (not by recount). Never as the sole basis for a trade.

Wyckoff (Richard Wyckoff, early 1900s)

The framework: a "Composite Man" (all large operators treated as one actor) accumulates in a range (Phases A–E), marks price up, distributes at the top, then marks price down. Three laws: supply & demand, effort vs result (volume vs price progress), cause & effect (the range builds the cause; the move out is the effect). Honest assessment: Wyckoff is a reading framework, not a forecasting system — its own serious practitioners say so. Volume is central (effort vs result is a genuine analytical idea), but phases are often only identifiable in hindsight, no win rate is claimed for Spring/Upthrust setups by credible sources (UNVERIFIED as a standalone edge), and forcing every range into the five-phase schematic is the classic beginner error. Where it earns its keep: teaching traders to read ranges with volume instead of chasing breakouts blind — with invalidation set before entry.

Modern retail methods: SMC/ICT, market profile, order flow

Smart Money Concepts / ICT

The vocabulary: order blocks, fair value gaps / imbalances (three-candle gaps price "must" return to fill), liquidity sweeps (stop raids above swing highs / below swing lows), breaks of market structure, mitigation/breaker blocks, premium/discount zones, kill zones. Verifiable: stops do cluster above swing highs and below swing lows, and breaks of those levels create predictable short-term order flow (Osler's order-book evidence is the mechanism). Claimed but not verified: that order blocks and FVGs contain resting institutional orders; that FVGs fill at the 80–90% rates quoted online (no verified fill-rate study exists — those figures are unsourced lore).

The independent test (2026). Tangirala (Sep 2026, preprint, not yet peer-reviewed) formalised two widely used open-source SMC/ICT implementations into 44 detectors and tested them on 496 S&P 500 stocks, daily bars, Jan 2010–Jun 2026: 2.20 million events, 17.6 million simulated trades, each measured against its own ticker's unconditional return with multiple-testing control. Result: against the matched null, none of the 44 concepts beat random entry at any horizon; the largest point estimates (11–12 bp over ten days) had p ≥ 0.23; concepts selected on three years of data earned −31 bp out-of-sample. One preprint, one market, daily bars — not a final verdict on every SMC variant — but it is the only large-scale independent audit that exists, and it found nothing. Treat SMC/ICT as unproven; its component ideas (structure, levels, stop clusters) are sound, but they are classical technical analysis with new names.

Market profile & auction theory

Markets are a continuous two-way auction seeking a price where buyers and sellers agree. Market Profile organises the session as TPOs (one letter per 30-minute period at each price level), showing the Point of Control, the Value Area (~70% of activity), and the Initial Balance. The descriptive power is genuine — a profile shows where the market accepted value vs rejected price. But the framework's rules of thumb (e.g. the "80% rule") are practitioner observations from the CBOT pits, not published statistical results (UNVERIFIED as a probability). No peer-reviewed study proving a standalone market-profile trading edge was found. Use as session context combined with an entry method and risk rules, not as a signal generator. Profiles work best on liquid, session-structured instruments (Nifty/Bank Nifty futures, liquid largecaps); on illiquid names the profile is mostly noise.

Order flow / footprint charts

Footprint charts show, per price level within each candle, how much volume traded at bid vs ask (delta), plus imbalances and absorption. The data shown is real, and concepts like absorption (heavy volume, little price progress) and delta divergence are legitimate readings. But Indian equity order-flow data at retail level is limited (exchange data is aggregated; true depth beyond 5 levels is not available to most retail traders — verify current availability before paying for tools), and no published study demonstrates a standalone footprint-trading edge (UNVERIFIED). Footprint is a confirmation and context tool, not a signal generator.

Multi-timeframe analysis

Read the higher timeframe for trend/bias, the middle timeframe for structure, the lower timeframe for entry timing (e.g. daily → 1-hour → 15-minute). Rules that prevent contradiction: one job per timeframe; two timeframes agreeing is not "double confirmation" — they share the same price history, so agreement means "no contradiction," nothing more; levels cascade down, not up; when timeframes conflict, the higher one wins for direction. Multi-timeframe analysis is practitioner convention with no formal published test as a standalone method (UNVERIFIED). Its value is organisational — it stops traders from taking every lower-timeframe signal blind — a real behavioural benefit, not a statistical edge.

Evidence scorecard

Ratings: Strong replicated, peer-reviewed, post-cost evidence. Moderate real evidence with important caveats. Weak mixed or small-magnitude evidence, or mechanism without outcome evidence. Unproven no rigorous evidence found either way.

Method	Rating	Reason (one line)
Trend / time-series momentum (multi-month)	Strong	Replicated across 58 markets and a century of data (MOP 2012; Hurst et al. 2017); crashes and costs are the caveats
Moving-average trend filter (risk control)	Moderate	Timing rules cut drawdowns historically (Faber 2007, secondary); crossover entries have no robust edge
Support/resistance zones (prior reaction points)	Moderate	Real mechanism (order clustering at round numbers, Osler) + universal use; exact levels are subjective
Round-number levels	Moderate	Documented order clustering (8.7% at 00 vs 1% expected); effect decayed over time in stocks
VWAP (intraday bias / execution benchmark)	Moderate	Documented institutional mechanism; standalone retail edge unproven
Volume profile / market profile (context)	Weak	Genuine descriptive power; "80% rule" and edge claims are unverified practitioner lore
ATR (sizing / stops)	Moderate	Not a signal — a volatility unit; its use in sizing is arithmetic, not a claim
RSI / Stochastic / CCI (standalone)	Weak	Effects are market- and parameter-specific; no robust post-cost standalone edge
MACD (standalone)	Weak	Win rate under 50% alone in one US study; trend-confirmation use is defensible
Bollinger Bands (mean reversion)	Weak	Volatility measurement is real; band-touch reversal claims are folklore; one Nifty study is unverified
Chart patterns (H&S, double tops, triangles)	Weak	Lo et al.: incremental information, small magnitudes, gone after costs; Osler: H&S trading unprofitable before costs
Candlestick patterns (standalone)	Weak	Marshall et al. 2006: no value on DJIA; Caginalp 1998 positive but old/small sample; some 3-day patterns profitable only with strict rules (Lu et al. 2015)
Heikin Ashi / Renko / P&F	Weak	Noise filters with lag and distorted prices; no profitability evidence; trend filter only
Fibonacci retracement/extension	Unproven	Backtests find no difference vs random levels; self-fulfilling case needs identical grids that don't exist
Elliott Wave	Unproven	Unfalsifiable in real time (re-countable after any move); no peer-reviewed validation
Wyckoff	Unproven	Reading framework with real volume logic; phases often clear only in hindsight; no win rate proven
SMC / ICT (order blocks, FVGs, sweeps)	Unproven	Only large independent test (Tangirala 2026, preprint): none of 44 detectors beat random entry; components are classical TA renamed
Order flow / footprint (standalone)	Unproven	Real data, legitimate readings; no published standalone edge; Indian retail data access is limited
Multi-timeframe analysis	Unproven	Sound organisation and behavioural benefit; no formal test as a method
ADX / Parabolic SAR / OBV / A-D line	Unproven	No rigorous profitability evidence found; regime-filter and stop-placement uses are logic, not evidence

The pattern in the scorecard is the lesson

The methods with the strongest evidence are the least glamorous (follow the multi-month trend, respect levels where orders cluster, size by volatility, use VWAP for intraday context). The methods with the most YouTube views (Elliott, ICT, Fibonacci targets) have the least evidence. That inversion is not a coincidence — simple, testable ideas get tested and found wanting or modest; untestable ideas survive because they cannot be killed.