Technical Analysis of the Financial Markets — John J. Murphy
Macro Overview & Strategic Value
Chapter 9 makes the pivotal transition from subjective, discretionary chart reading to fully quantified, mechanical trend-following. Murphy’s core thesis is that the moving average’s greatest strength — its precise, unambiguous, computer-programmable signal generation — is what makes it the backbone of virtually every systematic trend-following system in existence, in sharp contrast to pattern recognition, which two technicians can legitimately disagree about.
This matters to a practitioner because it’s the book’s first fully backtestable, rules-based methodology: entry, exit, and even risk-adjustment logic can be reduced to precise crossover conditions rather than discretionary judgment. Every choice covered here — average type, length, number of averages, envelope width — represents an explicit sensitivity-versus-reliability trade-off a system designer must resolve, making this chapter the direct conceptual predecessor of quantitative, computer-driven CTA and managed-futures methodologies.
The chapter closes by candidly stating the moving average’s core limitation — it only performs well in trending markets, roughly a third to half of the time — which structurally sets up the next chapter’s introduction of oscillators as the necessary complementary tool for trendless conditions, and previews the ADX indicator (Chapter 15) as the mechanism for knowing which tool regime to apply.
Core Concepts & Mechanics
- Moving average as lagging, not predictive — It only reacts to price after a trend is already underway; it’s explicitly described as “a follower, not a leader,” meaning a trader accepts inherent signal delay in exchange for objectivity.
- Sensitivity/reliability trade-off by length — Shorter averages hug price closely and generate earlier but more frequent false (whipsaw) signals; longer averages lag more but filter out noise — a direct parameter choice a trader must calibrate to the market’s trending versus choppy character.
- Simple vs. linearly weighted vs. exponential averages — Simple averages weight all included days equally and drop older data entirely once outside the window; linearly weighted averages overweight recent days but still discard older data; exponentially smoothed averages weight recent data more heavily while mathematically retaining influence from the entire price history.
- Single moving average crossover — A buy/sell signal triggers when price crosses above/below the average; adding the condition that the average itself must also turn in the crossing direction filters out weaker signals.
- Double crossover method — A buy signal fires when a shorter average crosses above a longer one (e.g., 5/20 or 10/50); this reduces whipsaws versus a single average at the cost of slightly more lag, since two lines must align rather than just price versus one line.
- Triple crossover (4-9-18 system) — Requires proper stacking of three averages (shortest above middle above longest in an uptrend); a crossing of the fastest average generates an early “alert,” while the middle average crossing the longest gives full “confirmation” — providing a staged, two-tier signal-confidence framework.
- Envelopes and Bollinger Bands — Fixed-percentage envelopes (e.g., 3% around a 21-day average) flag overextension at constant width, while Bollinger Bands (2 standard deviations around a 20-day average) dynamically expand/contract with volatility, making narrow bands a precursor to new-trend initiation and wide bands a precursor to trend exhaustion.
- The 4-week rule (Donchian channel breakout) — Buy when price exceeds the prior 4 calendar weeks’ high, sell/short when it falls below the prior 4 weeks’ low; a fully mechanical, non-discretionary breakout system empirically shown (per cited Merrill Lynch and Lukac studies) to be among the most robust trend-following approaches tested.
- Adaptive Moving Average (AMA) — Uses an Efficiency Ratio (comparing net price direction to total volatility) to automatically speed up in trending conditions and slow down in choppy conditions, solving the fixed-length average’s regime-dependence problem algorithmically rather than through manual parameter selection.
Technical Terminology & Reference Table
| Term | Operational Definition |
|---|---|
| Simple moving average (SMA) | Arithmetic mean of the last N closing prices, equally weighted |
| Linearly weighted moving average | Average giving progressively heavier weight to more recent days |
| Exponentially smoothed average | Weighted average retaining all historical data with recent days weighted most |
| Double crossover | Buy/sell signal from a shorter average crossing a longer average |
| Triple crossover (4-9-18) | Three-average system requiring alert + confirmation stacking order |
| Envelope | Fixed-percentage band around a moving average marking overbought/oversold |
| Bollinger Bands | Bands set at ±2 standard deviations around (typically) a 20-day average |
| Standard deviation | Statistical dispersion measure; ±2 SD captures ~95% of price data |
| Whipsaw | False signal generated by a moving average crossover in a trendless market |
| 4-week rule (Donchian channel) | Breakout system: buy above prior 4-week high, sell below prior 4-week low |
| Efficiency Ratio | Kaufman’s ratio of net price direction to total volatility, driving the AMA’s speed |
| Harmonics (cyclic) | Principle that neighboring time cycles relate by a factor of 2 |
The Author’s Market Philosophy
Murphy’s model of edge here is mechanical rather than interpretive — he explicitly contrasts the moving average’s virtue (precise, non-debatable signals) against the subjectivity of pattern-based chart reading, revealing a belief that removing discretionary judgment reduces error and enforces discipline (“let profits run, cut losses short”) more reliably than human interpretation alone. His view of market behavior is explicitly regime-dependent: markets alternate between trending and trendless conditions roughly evenly, and no single tool — however well-designed — works across both regimes, which is why he treats moving averages as necessarily complementary to (not a replacement for) oscillators. On optimization, his stance is skeptical rather than dismissive: parameter-fitting is not “the Holy Grail,” and he distinguishes disciplined out-of-sample testing from data-mined overfitting, reflecting an assumption that consistent process (not a perfectly-tuned parameter) is the more durable source of edge.
Systemic & Portfolio Integration
The double/triple crossover systems and the Donchian 4-week rule are presented as direct blueprints for systematic, mechanical trend-following and CTA/managed-futures methodologies, with cited empirical studies showing channel breakout and moving-average crossover systems outperforming other tested approaches historically. Bollinger Band width is explicitly framed as a volatility-regime signal feeding into risk management (position sizing/stop placement should differ between expanding and contracting volatility), while the chapter’s honest acknowledgment that trend-following underperforms roughly a third to half of the time is a direct expectancy input for blending trend and oscillator-based systems within a broader portfolio.
Important Formulas, Data, or Initial Examples
- Common daily crossover combinations: 4/9/18, 9/18, 5/20, 10/50 (futures); 50-day or 10-week single average (stocks).
- Envelope example: 3% envelope around a 21-day average (short-term); 5% envelope around a 10-week average or 10% around a 40-week average (longer-range).
- Bollinger Band construction: 20-day moving average ± 2 standard deviations, capturing ~95% of price observations within the bands.
- 4-week rule example: buy/cover shorts when price exceeds the high of the preceding 4 full calendar weeks; sell/short when price falls below the low of the preceding 4 weeks; can be shortened to a 1-2 week rule for exits or lengthened to 8 weeks for choppier markets.
- Cyclic harmonic logic: monthly cycle (~20-21 trading days) explains popularity of 5/10/20/40-day averages and 1/2/4/8-week rule variants, each related to neighbors by a factor of 2.
- Historical validation: Merrill Lynch studies (1978-82) and Lukac et al./Lukac & Brorsen research (1975-86, 23 systems tested) found channel breakout and moving-average crossover systems consistently outperformed other tested mechanical systems.
Active Recall Evaluation
- Explain the specific trade-off Murphy identifies between using a shorter versus a longer moving average, and why he adds the corollary that longer averages “give back” more when a trend reverses.
- Why does the triple crossover (4-9-18) system distinguish between an “alert” and a “confirmed” signal, and what does that structure protect a trader against?
- How does Bollinger Band width function differently from a fixed-percentage envelope, and what does an unusually narrow band typically precede?
- What is the core mechanical problem the Adaptive Moving Average’s Efficiency Ratio is designed to solve, and how does it decide when to speed up or slow down?
- Why does Murphy advise traders who follow only a handful of markets to consider optimization, while advising generalists who follow many markets against it?
Answer Key (spoiler)
- A shorter average hugs price closely and generates earlier trend signals, which is valuable for early participation, but its sensitivity also produces more false whipsaw signals during noisy or sideways action; a longer average filters out that noise and performs better while a trend remains intact, but because it trails price from a greater distance, it “gives back” more of the accumulated profit before it finally registers a trend reversal — the same insensitivity that helps during the trend becomes a liability at the turn.
- The fastest (4-day) average crossing the other two only reflects the most sensitive, most noise-prone data reacting first, so treating it as a mere “alert” avoids overcommitting capital to a signal that could still be noise; requiring the 9-day to also cross the 18-day before confirming the trade adds a second, less sensitive layer of agreement, protecting the trader against acting on a premature or false signal from the fastest line alone.
- A fixed envelope maintains constant percentage width regardless of market conditions, while Bollinger Bands are built from a rolling standard deviation, so their width mechanically expands during high-volatility periods and contracts during low-volatility periods; an unusually narrow band signals compressed volatility, which Murphy notes often precedes the initiation of a new significant price move once volatility expands again.
- The core problem is that a single fixed-speed moving average cannot simultaneously perform well in both trending and choppy/sideways conditions — a fast average whipsaws in a range, a slow one lags too much in a trend. The Efficiency Ratio compares net directional price movement to the total volatility traveled over the same period; when directional movement dominates (high ratio), the AMA speeds up to track the trend more closely, and when volatility dominates with little net direction (low ratio), it slows down to avoid reacting to noise.
- Following only a handful of markets makes it practical and worthwhile to fine-tune parameters to each specific market’s unique behavior (e.g., Treasury Bonds shouldn’t necessarily use the same average length as corn), since the trader has the bandwidth to properly validate those optimized settings out-of-sample; a generalist tracking thousands of stocks or many markets cannot realistically optimize and validate parameters individually for each one, making a single consistent, non-optimized parameter set the more practical and robust choice across a large portfolio.