Side Lesson 23: Technical Analysis — What Works (Some), What Doesn't (Most), and the Academic Verdict

---

1. Why This Is Important

Technical analysis is the largest cottage industry in finance. There are tens of thousands of books, hundreds of indicators with three-letter acronyms, dozens of YouTube channels selling pattern-recognition courses, and a small army of CMTs (Chartered Market Technicians) whose certification body has been around since 1973. Against all this, the academic literature has been remarkably consistent for forty years: most of it doesn't work, two narrow pieces of it do. Side 23 is the honest survey of which is which.

**The "TA works" claim is mostly survivorship bias and

pattern-fitting.** Andrew Lo, Jegadeesh, Pedersen, Bessembinder and others have run rigorous tests on the universe of "classical" chart patterns — head-and-shoulders, double tops, triangles, pennants, flags — and the result is consistent: after transaction costs, no systematic edge. The patterns are visible after the fact because the brain is a pattern-matching machine. They are not predictive in advance.

But two pieces of TA do survive scrutiny. Time-series

momentum (the past 12 months of a security's own returns predicts the next 1-12 months) is documented in Moskowitz, Ooi and Pedersen 2012 across 58 markets and is the engine behind the $300B managed-futures industry (Week 51). The 200-day moving average as a risk-on / risk-off filter (Faber 2007) gives roughly the same long-run return as buy-and-hold but cuts the worst drawdowns roughly in half. These are not flashy. They are the exceptions, not the rule.

**The behavioural use is real even when the predictive use

isn't.** A trader who says "I sell when the 50-day breaks the 200-day" has, at minimum, a rule — a pre-committed binary trigger that overrides loss aversion (Side 15). The trigger may have weak predictive power, but the discipline is real. The market can stay irrational longer than you can stay solvent. A bad rule consistently followed beats a good thesis abandoned in a panic.

You will see this stuff every day on financial media. CNBC

guests will draw trendlines on screen. X / Twitter accounts will post "this triangle is breaking out." YouTube channels will sell you a course on Elliott Wave theory for $1,997. You need to be able to tell the difference between the 95% that is decoration and the 5% that is real, especially because the decorative kind tends to come dressed in the language of the real kind.

This lesson covers chart-pattern claims (the failures), the two TA strategies that replicate out-of-sample (the survivors), and the behavioural-discipline argument for using TA-style rules even when their predictive value is weak. The two anchors: alpha is rare, so the default should be passive; and momentum + mean-reversion + vol-on / vol-off are the only short-list of robust market regularities.

The interactive lab on the website lets you slide the moving-average window from 50 to 365 days and pick the risk-off asset (T-bills, gold, or 0%-cash) and see CAGR, vol, Sharpe and max drawdown vs buy-and-hold across the whole 1990-Apr 2026 sample.

---

2. What You Need to Know

2.1 What Technical Analysis Actually Claims

Technical analysis is the use of past price and volume data — and only past price and volume data — to predict future returns. The underlying claims are usually some combination of:

Price discounts all known information (a strong-form efficient-

markets statement, oddly enough).

Prices move in trends, and trends persist.

History repeats because human psychology is constant — the same

patterns of fear and greed produce the same chart shapes.

The first claim is identical to what passive indexers believe and implies you can't beat the market by any method. The second and third claims are testable empirical statements. They have been tested. The results are mixed at best.

The TA universe splits roughly into four buckets:

• Chart patterns — head-and-shoulders, double tops / bottoms,

triangles, pennants, flags, cup-and-handle, wedges. Visual shape-based.

• Indicators — RSI, MACD, stochastics, Bollinger Bands, ADX,

CCI, ATR. Mathematical transforms of price.

• Trend / momentum rules — moving-average crossovers, 12-month

return ranking, Donchian channels, breakout systems. Time-series trend-following.

• Esoteric — Elliott Waves, Gann angles, Fibonacci retracements,

candlestick patterns (doji / hammer / engulfing). Highly subjective.

The literature finds: bucket 3 has some real signal. Buckets 1, 2, and 4 do not, after costs.

2.2 The Failure: Chart Patterns and Indicators

The most rigorous chart-pattern study is Lo, Mamaysky and Wang (Foundations of Technical Analysis, Journal of Finance 2000). They wrote a kernel-regression algorithm to detect 10 classical patterns (head-and-shoulders, broadening tops, triangles, etc.) on US equities 1962-1996 — taking subjectivity out by formalising "what counts as a pattern" mathematically. Result: a few patterns showed weak statistical signals (head-and-shoulders had a non-zero difference of conditional vs unconditional return distributions), but the magnitudes were small and the differences disappeared after transaction costs. Subsequent out-of-sample tests (Bessembinder & Chan, Brock-Lakonishok-LeBaron extensions) found the weak signals didn't replicate post-1996.

Indicators have fared no better. Park & Irwin's 2007 survey of 95 academic studies on technical trading found that 56 reported positive results, 20 negative, 19 mixed — but the positive-result studies were heavily weighted toward FX and commodities pre-1990 (when frictions were huge and the signal was almost certainly cost- related), and the failure rate climbed sharply for post-1990 equity samples. By the 2010s, the consensus view was that classical indicators (RSI, MACD, stochastics) had no robust out-of-sample edge in liquid US equities.

Candlestick patterns — Marshall, Young & Rose 2006, then Horton 2009 — were tested on the full Japanese-defined set across US large- caps and showed no profits net of costs. The "doji means indecision" claim turns out to mean roughly the same thing as "a coin flip means indecision." It does, but it doesn't help you.

Elliott Waves and Fibonacci retracements have effectively zero academic support. The Elliott Wave principle requires you to identify the current "wave count" — but the methodology allows re-counting after any reversal, which means it cannot be falsified in real time. The Fibonacci retracement levels (38.2%, 50%, 61.8%) are tested in Bhattacharya & Kumar 2006 and others; result, in the authors' words: "no statistically significant predictive content."

2.3 The Survivor #1: Time-Series Momentum (Moskowitz-Ooi-Pedersen)

The first TA-adjacent finding that has replicated robustly across markets and decades is time-series momentum — the observation that *a security's own past 12-month return predicts the next 1-12 months of returns, on average, in the same direction*. Moskowitz, Ooi and Pedersen's 2012 paper Time Series Momentum tested 58 liquid futures markets (equity indexes, currencies, commodities, bonds) from 1985-2009 and found that buying assets up over the last 12 months and shorting assets down over the last 12 months produced a Sharpe ratio of roughly 1.0-1.2 at the diversified-portfolio level — far higher than the underlying assets in isolation.

This is the engine that drives the managed-futures / CTA industry (Week 51). DBMF, KMLM, FMF, AHLT all run some flavour of time-series trend on a diversified futures basket. The 2008 +13.1% and 2022 +20.5% returns of the SocGen CTA index were essentially the time- series momentum signal capturing extended trends in rates, FX and equities while everything else was selling off.

Time-series momentum is a TA strategy in the strict sense — the only input is past prices. But it is the opposite of "head-and-shoulders on a daily chart." It works on long horizons (3-12 months), across a diversified basket (not single stocks), with explicit rules (rank-by-12-month-return-and-rebalance-monthly), and almost certainly captures a behavioural under-reaction to information that takes time to be priced in. Momentum is one of the two robust market regularities. This is where it shows up.

2.4 The Survivor #2: The 200-Day Moving Average (Faber GTAA)

The second survivor is the **200-day moving average as a risk-on / risk-off filter**, popularised by Mebane Faber's 2007 SSRN paper A Quantitative Approach to Tactical Asset Allocation (the most- downloaded paper in SSRN history at one point).

The rule:

• At month-end, check whether SPY's price is above its 10-month

(~200-day) simple moving average.

• If yes, hold SPY.
• If no, hold 3-month T-bills.

That's it. No discretion, no filters, no overlays.

The result, on US large-caps 1990-Apr 2026 (the exact backtest in the image script for this lesson, using ^GSPC price-only AdjClose for the equity leg and 3-month T-bills via FRED DGS3MO when out):

• Buy-and-hold S&P 500: ~8.6% CAGR, max drawdown -57% (2009).
• Faber 200-DMA rule: ~9.0% CAGR, max drawdown -24%.
• Sharpe ratio: ~0.32 buy-hold vs ~0.52 Faber.

(Adding the ~1.8 pp/yr SP500 dividend yield brings buy-and-hold's total-return CAGR closer to ~10.4%. The rule's CAGR rises in similar proportion when applied to a dividend-reinvested ETF; the *gap on max-drawdown* is what matters and is robust either way.) The CAGR is in the same ballpark; the maximum drawdown is less than half the size. The risk-adjusted return is meaningfully better.

This is the only piece of "technical analysis" most academics will defend in print. It is robust across decades, across markets (it works on EAFE and EEM too), and across reasonable parameter choices (any MA window from 100 to 250 days gives similar results — the result is not parameter-fitted to "200" specifically).

What it does not do: produce excess return in a long bull market. From 2010-2019, buy-and-hold beat the 200-DMA rule by ~1.5 pp/yr because the rule sat in T-bills during the August 2011, August 2015, February 2016 and December 2018 false breaks. The rule pays for itself in the once-or-twice-per-decade event where it cuts a -55% into a -20%. Vol-tail-wags-dog is the framing. You're not paying for upside; you're paying to truncate the left tail.

2.5 Why It Works: The Behavioural Story

The cleanest behavioural explanation for both survivors is *under- reaction*. New information takes time to be priced in — partly because attention is limited (Hong & Stein 1999), partly because disposition-effect selling caps gains early (Frazzini 2006), partly because institutional flows are slow (rebalance lag, mandate constraints, mutual-fund cash buffers).

When information is being slowly priced in, prices trend. A stock that beat earnings two months ago is still being upgraded by sell- side analysts, still being added to growth-strategy mandates, still being chased by retail. A stock that missed earnings two months ago is still being downgraded, dropped, sold. Time-series momentum and the 200-DMA both ride that trend.

When the information is fully digested, prices stop trending — and mean-reversion takes over (the other robust market regularity). This is why the trend rules whipsaw in flat / choppy markets and shine in directional ones.

The contrast with chart patterns is instructive. A "double top" is a shape, not a flow. There is no behavioural story under it — no mechanism by which the visual symmetry of the high produces a predictive signal. The pattern catches the brain's eye; it does not catch institutional capital. That is why it doesn't replicate.

2.6 The Behavioural Use Case: TA as Discipline

Here is the honest case for TA even where its predictive value is weak.

A retail investor with no rules tends to do the following:

• Buy when the stock is ripping (recency / FOMO from Side 15).
• Hold when the stock is down 30%, hoping for a bounce (loss-

aversion / disposition).

• Sell at the bottom when the news is darkest (capitulation, cf.

Dalbar 1.5-2 pp/yr behaviour gap).

A retail investor who follows even a weak technical rule — "I sell when the 50-day breaks the 200-day, no exceptions; I buy back when the 50-day crosses back above" — does the following:

• Sells in the early innings of major drawdowns (the rule fires in

Aug 2008, Feb 2020, Jan 2022 — early-to-middle of each crash).

• Buys back partway through the recovery (June 2009, May 2020,

Mar 2023).

• Has zero discretion in the moment, which is the moment when

discretion is worst.

The technical rule may be only weakly predictive, but it is *binary and pre-committed* — the two things that defeat behavioural bias (Side 15 §2.5). A bad rule consistently followed beats a good discretionary thesis abandoned at the bottom. The market can stay irrational longer than you can stay solvent — the rule is your insurance against your own brain when both you and the market are tilted.

This is also why the tactical-allocation industry ($150B+ AUM in 200-DMA-style rules) has staying power despite the modest CAGR edge. Clients are buying the behaviour insurance, not the return enhancement.

2.7 The Honest Verdict

Putting the literature together:

Bucket	Verdict	Notes
Time-series 12-mo momentum (futures basket)	Works	MOP 2012, Sharpe ~1.0 diversified, basis of $300B CTA industry
200-DMA / 10-month MA risk-on/off	Works (modestly)	Faber 2007 — same return, half the max drawdown
Cross-sectional momentum (single stocks)	Works (academic)	Jegadeesh-Titman 1993; -45% crash 2009; -1 pp/yr decay since
Chart patterns (H&S, double tops, triangles)	No edge	Lo-Mamaysky-Wang 2000; replicated post-2000
RSI / MACD / stochastics	No edge	Park-Irwin 2007 survey
Candlestick patterns	No edge	Marshall-Young-Rose 2006
Elliott Waves / Gann / Fibonacci	No evidence	Unfalsifiable in real time
TA as discipline mechanism	Real value	Behavioural insurance — beats discretion in crisis

If you are going to use technical analysis, use it for what works: trend-following on diversified baskets, 200-DMA on broad indexes, and as a rule-set to defeat your own behavioural biases. Skip the patterns and the indicator soup. They are decoration.

2.8 The Post-COVID Regime: Why the Vol Surface Beats the Chart at Size

One thing the rest of this lesson does not say loudly enough. Classical technical analysis — every chart pattern, every indicator in the textbook, the entire candlestick canon — was built in a market where the options book was a small derivative of the cash equity. That market is gone. Since roughly 2020, zero-days-to-expiry options have come to dominate intraday flow on the major US indices, and a meaningful share of what looks like "price action" on a chart is really dealer hedging of options exposure pushing the underlying around. The option tail wags the equity dog. Reading the chart in the post-COVID regime without the corresponding option flow is reading the surface effect without the cause, and the older the pattern in your toolkit, the weaker its signal in the new regime.

The implication is uncomfortable for chartists, and I say this as someone who used to draw the lines: implied volatility, skew, term structure, and the dealer gamma profile are now more informative than the delta of the price itself. The vol surface tells you who has to do what at which level, regardless of whether the positioning is smart or dumb. The chart shows you the dog wagging. The vol surface shows you why. Side 20 is the deep dive on Greeks and the surface; this paragraph is the pointer that tells you when to go read it.

The cost-benefit is size-dependent, and that matters because most readers of this course are not yet at the size where the new toolkit pays for itself. For a beginner with their first portfolio, the cognitive overhead of vol-surface analysis is real and not yet earning its keep — start with the chart, run the 200-DMA rule, let the discipline do the work. For the serious retail investor with a multi-six-figure or seven-figure book, where individual position outcomes meaningfully change the year and not just the month, ignoring the vol surface stops being a small handicap and becomes a structural blind spot. Below that threshold, the chart still earns its keep — mostly as a behavioural anchor, the binary-and-pre-committed thing that defeats your own brain in March 2020. Above it, the chart is the appetiser and the surface is the main course.

---

3. Common Misconceptions

"All technical analysis is bunk." Not quite. Time-series

momentum and the 200-DMA filter both replicate out-of-sample across decades and markets. Two narrow pieces of it work. The other 95% does not. The honest position is "mostly bunk, two exceptions" — not blanket dismissal.

**"Technical analysis works because everyone watches it — self-

fulfilling prophecy."** This is the most common defence and it doesn't hold. If a level were truly self-fulfilling, sophisticated traders would front-run it, eliminating the edge. The empirical tests show no edge net of costs at the levels traders actually watch (50-day, 200-day, prior highs / lows). The story is appealing; the data don't confirm it.

"Head-and-shoulders is reliable — I see it work all the time."

You see it work because of confirmation bias (Side 15). When the pattern works, you remember it. When it doesn't, you say "that wasn't a real head-and-shoulders." Lo-Mamaysky-Wang formalised pattern definition exactly to defeat this — and the edge disappeared.

"RSI < 30 means oversold and the stock will bounce." RSI

tested across US equities since 1990 shows no statistically significant edge after costs. RSI < 30 in a downtrend tends to stay < 30. The "bounce" is the brain pattern-matching to a few high-profile cases.

**"Moving averages don't work because by the time the signal

fires, the move is over."** Wrong empirically. The Faber 200-DMA rule fired in October 2008 (-25% from peak, before the worst -30% to come), in March 2020 (-15% from peak before the -34% bottom), and in January 2022 (-5% from peak before the -25% trough). The rule got out early, not late.

"Trend-following only works in trending markets." True, but

trending markets are the ones where buy-and-hold has the largest drawdowns. The trend rule's job is exactly to capture the down- trends that buy-and-hold can't avoid. It will lose 1-2 pp/yr in choppy bull markets — that's the premium you pay for the left-tail insurance.

"Cross-sectional momentum (Jegadeesh-Titman) is dead." Not

dead, but decayed. The premium has compressed since the 2003 academic-publication date (McLean-Pontiff 2016, ~50% post-pub decay). The 2009 momentum crash was -45% in one quarter as beaten-up cyclicals snapped back. Single-stock momentum is real but expensive to harvest at retail (high turnover, big drawdowns, tax-inefficient). Time-series momentum on futures is more robust.

"Volume confirms price." Sometimes. The On-Balance-Volume

indicator (Granville 1963) has no statistical edge in modern US equities. Volume patterns around earnings and corporate events carry information, but generic "volume up on green days = buy" is not predictive after costs.

"Elliott Wave theory works if you count the waves correctly."

The unfalsifiability is the giveaway. Any methodology that lets you re-count after any move can be fitted to anything in hindsight. Robert Prechter (the most prominent Elliott Wave advocate) was in print calling for a Dow at 1,000-3,000 from the late 1990s through 2010s while the Dow went from 7,000 to 14,000 to 36,000.

"You can't backtest TA because the patterns are subjective."

Lo-Mamaysky-Wang's contribution was specifically to make the patterns objective via kernel regression. Once formalised, the patterns become testable. The tests showed no edge. Subjectivity is the defence of TA against falsification, not its strength.

---

4. Q&A Section

Q1. So is technical analysis worth learning at all? Two pieces of it: yes. (a) Time-series momentum on a diversified basket — read Moskowitz-Ooi-Pedersen 2012, then look at DBMF / KMLM as the implementation (Week 51). (b) 200-day moving average as a risk-on / risk-off filter on broad indexes — Faber 2007, the implementation in Side 23 §2.4 above. The rest — chart patterns, RSI / MACD soup, candlesticks, Elliott Waves — is decoration. Skip it.

**Q2. The 200-DMA rule has slightly lower CAGR than buy-and-hold. Why use it?** You're not buying CAGR. You're buying a left-tail insurance. The rule cuts max drawdown roughly in half (~-19% vs ~-55% in 2008-09). For an investor close to retirement or one whose behaviour falls apart at -40%, that left-tail truncation is worth the 1-1.5 pp/yr of foregone upside. Vol-tail-wags-dog is the framing.

**Q3. Is cross-sectional momentum (long winners / short losers among individual stocks) the same as time-series momentum?** Related but distinct. Cross-sectional ranks stocks against each other — buy the top decile by trailing 12-month return, short the bottom. Time-series ranks each asset against itself — long when its own 12-month return is positive, short when negative. The cross-sectional version had a -45% crash in Q1-Q2 2009 (the famous "momentum crash"), the time-series version on diversified futures did not. Time-series is more robust for retail / smaller funds.

**Q4. Why do so many CMTs and chartists make money if the methods don't work?** Several reasons. (1) Some run fundamentally and use TA only for entry-timing — the alpha is the fundamental call, not the chart. (2) Some use trend rules (which work, modestly) and call it "technical." (3) Survivorship — the ones who blew up don't have podcasts. (4) For floor traders pre-2000, the edge was *liquidity provision*, not pattern recognition; charts were the visible part of a market-making operation. (5) Some are simply selling courses; their P&L is the course revenue, not the trading.

Q5. Should I use stop-losses based on technical levels? On individual stocks, sized at 2-3 ATR (average true range) below entry, yes — they enforce position-level risk discipline. On index funds, no — indexes mean-revert, and stops on indexes typically get whipsawed into selling at the bottom. The right risk control on index sleeves is position sizing, not stops (Side 15 §3 #6).

Q6. Does the 200-DMA rule still work in 2026? Through Apr 2026 the answer is yes — the rule fired correctly out of the Jan 2022 -25% drawdown (sat in T-bills earning 4-5%, re- entered in Mar 2023), and through 2024-25 it stayed long. CAGR and max-drawdown on the rule remain materially better than buy-and-hold on max-drawdown and slightly worse on CAGR, consistent with the historical record. The interactive lab on the website lets you re-run the backtest and verify the numbers yourself.

**Q7. Why does the academic literature treat trend-following as "different from TA" when it's clearly TA?** Branding. Academics call trend-following "time-series momentum" because the term has clean econometric definitions and the literature runs through journals (Journal of Finance, Review of Financial Studies). Practitioners call the same thing "trend-following" and publish in trade journals. The mechanics — buy if rising, sell if falling — are identical. The literature is less hostile when it's called by its academic name.

**Q8. What about volume / OBV / accumulation-distribution indicators?** Volume around event moments (earnings, M&A, FDA approvals) is informative — it reflects information flow. Volume on quiet days filtered through indicator transforms (OBV, A/D line) shows no robust predictive edge. The question to ask of any volume claim: *what is the behavioural mechanism by which this signals price direction?* If there isn't one, treat the claim with scepticism.

Q9. Are chart patterns useful at all? Two narrow uses. (1) As a risk-management prompt — if a stock breaks decisively below a multi-year support level on heavy volume, that is information about institutional selling, even if the specific "head and shoulders" framing is decoration. (2) As a mental shorthand for trend and volatility regimes. Neither use generates alpha per se — they help you not lose money in obvious ways. The pattern-as-prediction claim is the part that doesn't replicate.

Q10. How does TA fit into the four-tranche framework? Tranches 1-3 (growth, income, stores of value) are passively allocated and benefit from no TA whatsoever — buy-and-hold the index funds and rebalance annually. Tranche 4 (opportunistic / 5% slot) is where TA-style rules can earn their keep — the 200-DMA overlay on the equity sleeve, time-series momentum exposure via DBMF or KMLM, or rule-based entry timing on individual positions. Confine TA to the small slice and you can't blow up the whole portfolio with a bad chart read.

**Q11. What's the single rule a retail investor should adopt from this lesson?** Either none (just hold the index — default passive, alpha is rare), or one: the 200-day moving average filter on your equity sleeve, applied at month-end on the SP500 index price, with no discretionary overrides. That single rule cuts max-drawdown in half historically, costs ~1 pp/yr in foregone CAGR, and gives you a behavioural anchor in crisis. Anything beyond that is incremental at best, distracting at worst.

Q12. Where does this leave Elliott Waves and the CMT certification? Elliott Waves: zero academic support, unfalsifiable methodology, treat as financial astrology. The CMT (Chartered Market Technician) designation: the content covers the full TA universe including the bunk; the credential signals seriousness about market microstructure but not predictive skill per se. If you're hiring a CMT, ask them which of the four buckets in §2.1 they trade. If they say buckets 1, 2, or 4, run.