Technical Analysis for Futures Traders
Part II, Classical TA Chapter 7

Oscillators: RSI, Stochastics, and the Mean-Reversion Trap

Overbought in a trend extends. Overbought in a range fades. The oscillator does not tell you which.

7.0Why this chapter exists

Oscillators are bounded transformations of price momentum. RSI, Stochastics, CCI, Williams %R, all map recent price action into a fixed range (0 to 100, or −100 to +100), giving the visual impression of a calibrated thermometer. The interpretation is universal in retail TA: high readings = overbought = expect reversal down; low readings = oversold = expect reversal up.

The interpretation is wrong without context. In a trend regime, "overbought" is exactly the condition you would expect in a strong up-move; the price is rising faster than recent baseline, and it will continue to rise. Mean-reverting against the trend at "RSI > 70" is one of the most reliable ways to lose money in liquid futures.

The work below proceeds in three movements. First, the major oscillators get defined for what they actually compute, which is mostly the same information rendered four different ways. Second, the standard "overbought" and "oversold" reading gets dismantled and rebuilt as a conditional signal that depends entirely on regime, because the same RSI 75 reading is anti-information in a trend day and a fade trigger in a Range-Calm session. Third, divergence (classical and hidden) gets treated honestly, with the rule that divergences at structure are useful and divergences elsewhere are not.

The chapter is short by design. Like Chapter 6, the institutional treatment of oscillators is mostly about not using them the way retail TA does.

7.1The major oscillators

Relative Strength Index (RSI)

RS = average_gain(N) / average_loss(N)   (Wilder smoothing)
RSI = 100 − 100 / (1 + RS)

Where average_gain is the Wilder-smoothed average of positive price changes and average_loss is the symmetric of negative changes. The standard period is N = 14.

RSI captures the ratio of upward momentum to downward momentum, normalized to a [0, 100] band. Standard interpretation: RSI > 70 = "overbought," RSI < 30 = "oversold."

Stochastics

%K = 100 × (close − lowest_low(N)) / (highest_high(N) − lowest_low(N))
%D = SMA(3, %K)   (the slow stochastic; the fast stochastic uses %K directly)

Standard period N = 14. Stochastics compares the most recent close to the high-low range over the lookback window. Standard interpretation: %K > 80 = "overbought," %K < 20 = "oversold."

The interpretation is similar to RSI but the math is different. Stochastics is more sensitive to single-bar extremes (because the high-low range is bounded by the highest high and lowest low, both of which can be set by single bars).

Commodity Channel Index (CCI)

TP = (high + low + close) / 3   (typical price)
MA_TP = SMA(N, TP)
mean_dev = average |TP − MA_TP| over N bars
CCI = (TP − MA_TP) / (0.015 × mean_dev)

Standard period N = 20. Despite its origin in commodities, CCI is unbounded and frequently exceeds the canonical ±100 range during strong trends. Standard interpretation: CCI > +100 = "overbought," CCI < −100 = "oversold." More sophisticated practitioners use ±200.

Williams %R

%R = −100 × (highest_high(N) − close) / (highest_high(N) − lowest_low(N))

A scaling of stochastics into [−100, 0]. Standard period N = 14. Standard interpretation: %R > −20 = "overbought," %R < −80 = "oversold."

What they share

All four oscillators compute, in slightly different ways, the position of the current price relative to recent highs and lows or relative to a smoothed mean. They are bounded (or nearly bounded) and produce visually similar charts. Walking through the math reveals that they are not independent signals; they are correlated transformations of the same underlying price-momentum information. Stacking them on a chart provides redundancy, not orthogonal evidence.

7.2The mean-reversion trap

The standard retail framing: "RSI > 70, take a short, expect mean reversion."

Why this fails in trend regimes

In a trend, price consistently exceeds its recent mean in the trend's direction. RSI computed on that price will consistently read above 70 (in an up-trend) or below 30 (in a down-trend), often for dozens of bars in a row. A trader who shorts every "RSI > 70" reading in a Trend-Vol day takes 6 to 12 losing trades against the trend, with each loss compounded by the trend's continued extension while the trader holds.

Why this works in range regimes

In a range, price oscillates around a mean. RSI computed on that price oscillates between high and low extremes that mark the range boundaries. A trader who fades the high extreme and the low extreme captures the range structure mechanically. In Range-Calm regimes, the textbook oscillator usage works approximately as advertised.

The synthesis

The same oscillator signal has opposite implications in opposite regimes. The signal is not a signal; it is conditioning information. It says "price is at a recent extreme of momentum." Whether that extreme will reverse or extend depends on regime, not on the oscillator value.

This is the mean-reversion trap: treating oscillator readings as direction-of-trade signals rather than as conditioning information that needs regime context.

The fix

Make oscillator usage conditional on the regime classification from Chapter 2:

  • In Trend-Vol or Trend-Calm: "RSI > 70" or "RSI < 30" reading is anti-information about reversal. It is consistent with continuation and should not be used as a fade trigger. If anything, an "overbought" reading in a Trend-Vol up-trend is a confirming signal of the trend, not a reversal flag.
  • In Range-Calm: "RSI > 70" at structure (i.e., near the upper range edge) is a reasonable fade trigger. The oscillator is consistent with the range structure.
  • In Range-Vol: Oscillator extremes are noisy. The range boundaries are blurry, and RSI can spike to 80 or 20 during a noise-driven range expansion that does not reverse. Treat with caution; require additional confirmation.
  • In Squeeze: Oscillators in a squeeze tend to oscillate in a narrow band themselves. They are mostly uninformative; wait for a regime transition before reading them.

The conditional usage is the rule. The standalone "overbought / oversold" framing is the trap.

7.3Divergence: classical and hidden

Divergence is the most-cited oscillator pattern, and the one most likely to be useful when properly conditioned.

Classical divergence (reversal)

  • Bearish classical: price prints a higher high; oscillator prints a lower high. Interpretation: momentum is weakening even as price extends; reversal is plausible.
  • Bullish classical: price prints a lower low; oscillator prints a higher low. Interpretation: momentum is improving even as price drops; reversal up is plausible.

Hidden divergence (continuation)

  • Bullish hidden: in an up-trend, price prints a higher low; oscillator prints a lower low. Interpretation: pullback is shallow in price but deeper in momentum, suggesting trend continuation.
  • Bearish hidden: in a down-trend, price prints a lower high; oscillator prints a higher high. Interpretation: bounce is shallow in price, momentum-decay is consistent with continued down-trend.

Reliability of each

A consistent finding in practitioner backtests, and consistent with the mean-reversion-trap logic above:

  • Hidden divergence (continuation) is meaningfully more reliable than classical divergence (reversal). The reason: hidden divergence aligns with the regime-prevailing direction, whereas classical divergence requires the regime to change, which is the harder forecast.
  • Both types are most reliable at structure. Classical divergence at PDH or a naked POC outperforms classical divergence in the middle of a range. Hidden divergence at a structural retracement level (50% of the prior impulse, or AVWAP) outperforms hidden divergence at random pullback depths.

CVD vs RSI divergence

A practitioner-level observation: CVD divergence (Chapter 3, Chapter 11) is directly informative about order-flow intent, whereas RSI divergence is informative about price momentum derivatives. Where CVD divergence is available (any platform supporting bid/ask volume or BVC), it should be the primary divergence layer. RSI divergence is a fallback when CVD is unavailable, and a supporting confirm when both are.

The divergence template

A proper divergence-based trade requires:

  1. The divergence is at structure (high-score level from Chapter 5).
  2. The regime classification supports the divergence direction:
  3. Classical divergence (reversal): regime should be Range or transitioning. Anti-pattern in Trend-Vol.
  4. Hidden divergence (continuation): regime should be Trend. Anti-pattern in Range.
  5. Volume confirms (the swing where the divergence forms shows declining volume on the reversal swing, or expanding volume on the continuation swing).
  6. Order-flow agrees (CVD aligns with the divergence reading).

A divergence satisfying all four is a high-quality signal. A divergence with only one of the four is a flag, not a trigger.

7.4Failure swings: a structural variant of divergence

Wilder's original RSI work introduced the failure swing, a pattern that does not require comparison of price highs at all. A bearish failure swing on RSI:

  1. RSI rises into the overbought zone (above 70).
  2. RSI then declines below 70.
  3. RSI then attempts to rise but fails to exceed the prior overbought peak.
  4. RSI then breaks below the trough between the two RSI peaks.

The pattern is entirely on the RSI subpane. It does not require a divergence with price. The signal is a regime change in momentum.

Failure swings are less commonly used than classical divergence, but they have a cleaner definition (the four conditions are mechanical) and they capture a structural change in the oscillator that is informative about regime transition. In practice, they tend to lead price-based signals slightly, which makes them a useful early-warning tool.

The same conditioning rules apply: failure swings at structure are more informative than failure swings elsewhere, and they should be paired with regime classification before sizing.

7.5Multi-timeframe oscillator reading

A technique borrowed from longer-horizon technicians: read the oscillator at two timeframes, with the higher timeframe setting bias and the lower timeframe selecting entry.

The protocol

  1. Higher timeframe: check whether RSI on the higher timeframe (e.g., 1H if you trade 5-min) is in extreme territory or trending.
  2. Lower timeframe: if the higher timeframe RSI is in oversold territory and turning up, look for long entries on the lower timeframe.

What this captures

The higher-timeframe oscillator filters out trades against the longer trend. It is similar in spirit to the 200-day MA filter from Chapter 6 (a long-horizon directional bias) but is more sensitive (responds within hours, not weeks).

Limitations

  • Whipsaws at oscillator regime turns. When the higher-timeframe oscillator is near a threshold, the bias flips frequently. Use buffered thresholds (e.g., bias = "long" only if higher-timeframe RSI < 35, "short" only if > 65, "neutral" between).
  • Disagreement between oscillators. RSI and Stochastics on the same chart frequently disagree on whether the current bar is "extreme." Pick one as the primary; do not stack them and call agreement.

7.6Failure modes specific to oscillators

  1. Trend-day mean-reversion losses. As above, mean-reverting against the trend on oscillator extremes during Trend-Vol or Trend-Calm is the most expensive mistake. The fix is the regime filter.

  2. Lookback period overfitting. RSI(14) is the default; some traders search for "the best" RSI(N) for their contract. Differences between 12, 14, 16, 18, 20 are small; the search is curve-fitting. Use the default.

  3. Stacking redundant oscillators. RSI agreeing with Stochastics agreeing with CCI is not three independent confirmations; it is one confirmation plotted three ways. The visual reassurance is false.

  4. Oscillator on the wrong instrument. RSI on a continuous back-adjusted chart at a roll boundary can produce spurious extremes because the underlying contract changed. Either re-anchor or use unadjusted data for the oscillator computation.

  5. Crisis-volatility readings. During regime breakdowns, oscillators saturate at extremes for extended periods (RSI pinned above 80 or below 20). The signal is uninformative; structure has overwhelmed momentum. Wait for normalization.

  6. Numerical edge cases. RSI computed during a long sequence of identical or nearly-identical bars (lunch chop, low-volume drift) can produce unstable values (denominator near zero). Handle with care; a gap between bars or a near-zero true range can spike RSI to 0 or 100 spuriously.

7.7The integrated stack treatment

Where do oscillators fit in the institutional stack?

  • Not in the primary stack. Oscillators are not on the recommended layer list (§10 of the Research Summary).
  • Useful as a divergence layer at structure. When CVD is not available (rare on modern platforms), RSI or Stochastics divergence at a high-score level can serve as a substitute. CVD is preferred where available.
  • Useful as a regime sanity check. RSI behavior on a daily chart, plotted alongside the regime composite, can confirm or refute the composite's classification: an RSI that has been pinned above 60 for 30 sessions confirms a trend regime, regardless of what other indicators say.

The institutional rule: oscillators are decoration on a chart that does not have CVD or footprint; they are nothing on a chart that does.

7.8Diagram concepts referenced in this chapter

  • D7.1: Oscillator family side-by-side. Four sub-panels with RSI, Stochastics, CCI, and Williams %R on the same price chart. The reader should see the high correlation among them.
  • D7.2: Trend-day RSI saturation. A 5-min ES chart of an extreme trend day with RSI plotted, showing RSI pinned above 70 for the majority of the session. Annotation: every "short the overbought" entry would have lost money.
  • D7.3: Range-day RSI as a fade signal. The opposite case: a Range-Calm session with RSI oscillating between 30 and 70, hitting each extreme at the range boundary. Faded entries each marked.
  • D7.4: Classical vs hidden divergence side-by-side. Two panels of the same chart segment showing each divergence type, with the structural and regime conditioning visible.
  • D7.5: Failure swing on RSI. A clean example of the four-stage failure swing pattern, with each stage numbered.

7.10Exercises

Exercise 7.1: The mean-reversion trap audit. On NQ 5-min for the most recent ten Trend-Vol days, mark every RSI > 70 reading. Tabulate the subsequent 10-bar reaction. Compute the conditional reaction rate (price reversed by ≥ 0.5 × ATR within 10 bars). The rate should be at or below the unconditional base rate, confirming that "RSI > 70" in a trend is anti-information for mean reversion.

Exercise 7.2: Range-day RSI fade. Repeat Exercise 7.1 on ten Range-Calm days. The conditional reaction rate should be visibly higher, confirming the regime conditioning.

Exercise 7.3: Hidden vs classical divergence. Over a 30-session sample on ES 5-min, identify all classical and hidden divergences on RSI. Tag each with regime, structural conditioning (at high-score level or not), and outcome. Tabulate the conditional reaction rate by (type × regime × structure). Hidden divergence at structure in a Trend regime should have the highest rate.

Exercise 7.4: RSI vs CVD divergence. On a recent NQ session, find five cases where RSI and CVD agree on a divergence and five where they disagree. Outcome distribution for each. The hypothesis: agreement (both showing divergence) is more reliable than RSI alone.

Exercise 7.5: Multi-timeframe RSI overlay. On an ES chart, plot RSI(14) on both 5-min and 1H subpanes. For each 1H RSI extreme over a recent week, count the number of 5-min RSI extremes that occurred while the 1H was in the same direction. The relationship should be visually clear; quantify the lead-lag.

Next chapter: volatility, ATR, Bollinger Bands, and the squeeze, treating volatility as the unit of measurement underlying every other indicator's interpretation.