Buy the Panic: What 20 Million Bollinger Tests FoundE-mini S&P 500 FuturesCME_MINI_DL:ES1!EdgeToolsThis is the fifth study in the series. The first tested RSI across 26 million configurations and found zero Bonferroni-significant results. The second documented a small but genuine Turn of the Month anomaly. The third examined VWAP and produced the strongest result to date: 150,546 Bonferroni-significant results concentrated in mean reversion, with short signal edge of 0.89 percentage points. The fourth tested MACD across 14.3 million configurations and found 3,235 barely significant results, all in divergence, with edge too small to trade on. Four studies, 46 million tests, one finding that matters: volume-weighted mean reversion. This fifth study examines Bollinger Bands and, in the process, complicates the narrative we had been building. Across the first four papers, a clean thesis was emerging: indicators that transform price alone produce nothing, while indicators that incorporate additional data dimensions, volume or the calendar, capture real structure. RSI and MACD failed because they are linear and nonlinear transformations of closing prices. VWAP succeeded because it multiplies price by volume. Turn of the Month succeeded because it accesses institutional flow cycles. Bollinger Bands are, technically, a price-only indicator. No volume. No external data. Just closing prices, a moving average, and a standard deviation envelope. Under the thesis of the first four studies, Bollinger Bands should fail. They do not. We tested 19,765,587 parameter configurations across six strategies and 14 assets. 320,256 results survive Bonferroni correction. The band penetration strategy, where the close falls below the lower band, produces long signal edge of 1.22 percentage points, the strongest single-strategy finding in this entire series, including VWAP. That was not expected. The catch, and it refines rather than destroys the thesis, is that Bollinger Bands are not a simple price transformation. The standard deviation envelope captures a second-order property of price, namely volatility, which is a genuine market state variable with well-documented mean-reverting characteristics. A close below the lower band means price has reached an extreme relative to its own recent volatility regime. That is a fundamentally different statement from RSI saying a number dropped below 30, and the data treats it differently. The popular strategies, squeeze breakouts, band bounces, and middle band crossovers, fail completely. Mean reversion from extreme band violations succeeds. The pattern is the same as VWAP. The mechanism is different. The conclusion is the same. Abstract We test six common Bollinger Band trading strategies across 14 liquid ETFs spanning five asset categories. From 19,765,587 parameter configurations covering band periods from 5 to 200, standard deviation multipliers from 0.5 to 4.0, and holding periods from 1 to 252 trading days, we find 320,256 results surviving Bonferroni correction at alpha equal to 2.53 times ten to the negative ninth power. This exceeds the Bonferroni count from any previous study in this series, including VWAP. The aggregate long edge is positive 0.29 percentage points and the aggregate short edge is positive 0.17 percentage points. These averages conceal the real finding. Band penetration, where the close falls below the lower band, produces long edge of positive 1.22 percentage points with 7,131 Bonferroni-significant results. The parametric generalization through Percent B thresholds extends this to 287,260 significant results at positive 0.56 percentage points long edge. Together, mean reversion strategies account for 318,547 of the 320,256 total Bonferroni results, a concentration of 99.5 percent. The Bollinger Squeeze, the most widely taught strategy in retail education, produces exactly zero Bonferroni-significant results from 7,202,640 tests. The middle band crossover produces zero from 96,040 tests. Bollinger Bands contain genuine predictive information when treated as a volatility-relative mean reversion framework. They contain none when used as the momentum, breakout, or support and resistance tool that dominates retail practice. 1. Introduction John Bollinger developed the bands that bear his name in the early 1980s and formalized them in his 2001 book (Bollinger, 2001). The original concept was straightforward: place an envelope around a simple moving average, where the envelope width is determined by the standard deviation of price over the same lookback period. When volatility expands, the bands widen. When volatility contracts, they narrow. The bands adapt to market conditions rather than imposing fixed thresholds. Bollinger was explicit about what the bands were designed to do. They provide a relative definition of high and low prices. A close at the upper band is relatively high. A close at the lower band is relatively low. The word "relatively" is doing the work. A stock trading at the lower band during a calm market has reached a different kind of extreme than the same stock trading at the lower band during a crash. The bands encode this distinction. Fixed-threshold indicators like RSI do not. What Bollinger did not claim is that the bands constituted a trading system. In his book, he described the bands as a framework for generating questions, not answers. He introduced Percent B, the position of the close relative to the bands, as a quantification tool. He described bandwidth, the distance between the bands normalized by the middle band, as a measure of the volatility cycle. He suggested the squeeze, a period of narrow bandwidth, might precede a volatility expansion. At no point did he assert that touching the lower band means buy, that the squeeze predicts direction, or that the middle band acts as reliable support and resistance. The retail trading community adopted the bands with enthusiasm and none of the nuance. Online forums, courses, and video tutorials teach three primary strategies: buy when price touches the lower band (support), sell when it touches the upper band (resistance), and trade the squeeze breakout by entering when bandwidth narrows and price breaks out directionally. These strategies are presented with conviction despite the absence of systematic evidence. The few academic examinations of Bollinger Band profitability, such as the work surveyed in Park and Irwin (2007), find mixed results at best, with studies that account for transaction costs and multiple testing generally finding no economically significant edge. This study tests Bollinger Band strategies more thoroughly than they have been tested before: 19,765,587 configurations across every common strategy type. The results contradict the retail consensus on what works but reveal something the retail community is not looking for. 2. What Bollinger Bands measure The calculation starts with a simple moving average of the closing price: Middle Band = SMA(close, n) The upper and lower bands are placed at a fixed number of standard deviations above and below: Upper Band = SMA(close, n) + k * StdDev(close, n) Lower Band = SMA(close, n) - k * StdDev(close, n) where n is the lookback period and k is the deviation multiplier. The standard 20/2 parameterization uses a 20-period SMA with bands at two standard deviations. Two derived indicators complete the framework. Percent B measures where the current close sits relative to the bands: Percent B = (close - Lower Band) / (Upper Band - Lower Band) A Percent B of 1.0 means the close is at the upper band. A Percent B of 0.0 means the close is at the lower band. Values below zero indicate that the close has fallen below the lower band, the condition we call band penetration. Bandwidth measures how wide the bands are relative to the middle band: Bandwidth = (Upper Band - Lower Band) / Middle Band Low bandwidth indicates low volatility. The squeeze, the condition retail traders watch for, occurs when bandwidth reaches historically low levels. The mathematical structure is more interesting than it initially appears. RSI and MACD are transformations of closing prices that produce outputs correlated with recent price momentum. They repackage the same information in a different format. Bollinger Bands also use closing prices as their sole input, but the standard deviation calculation introduces a second-order statistic. Standard deviation measures the dispersion of returns, which is volatility. Volatility is not merely a different representation of price level or direction. It is a distinct market characteristic with its own dynamics. This distinction matters because volatility has a well-established tendency to cluster and mean-revert. Engle (1982) demonstrated that volatility exhibits serial dependence, meaning that periods of high volatility tend to be followed by high volatility, and periods of low volatility by low volatility, with eventual reversion to the long-run level. Bollinger Bands embed this property directly: the bands widen during volatile periods and narrow during calm ones. A close below the lower band during a high-volatility regime represents a more extreme event than the same close during a low-volatility regime, and the bands automatically adjust the threshold to reflect this. Whether this adaptive thresholding translates into predictive power for future returns is the empirical question we address. The theoretical argument is that when price reaches an extreme relative to its own recent volatility, the combination of price mean reversion and volatility mean reversion creates a temporary dislocation that corrects. The data will determine whether this argument holds. 3. Common Bollinger Band strategies We tested six strategies representing the dominant applications of Bollinger Bands in retail trading practice. The band touch strategy generates a long signal when the session low reaches or crosses below the lower band but the close recovers above it. This interprets the lower band as dynamic support: price wicks below and bounces. The short signal fires when the session high reaches the upper band but the close falls back below it. The band penetration strategy generates a long signal when the close itself falls below the lower band. The short signal fires when the close exceeds the upper band. This is more extreme than band touch: the close does not recover within the session. Retail interpretation varies. Some treat penetration as a sign that the move will continue. Mean reversion logic suggests the opposite: an extreme that tends to correct. The Percent B threshold strategy generalizes band penetration using the Percent B indicator. A long signal fires when Percent B falls below a specified threshold. A short signal fires when Percent B rises above one minus the threshold. Seven threshold values from 0.0 to 0.30 are tested. At threshold 0.0, this strategy is mathematically equivalent to band penetration. Higher thresholds generate signals from less extreme deviations. The squeeze breakout strategy identifies periods where bandwidth falls below a specified percentile of its rolling 252-day distribution, then generates a long signal when price crosses above the middle band during the squeeze, and a short signal when price crosses below. Five percentile thresholds from the 5th to the 25th are tested. This is the strategy retail traders associate most closely with Bollinger Bands, based on the idea that compressed volatility precedes directional moves. The trend following strategy generates a long signal when the close exceeds the middle band and a short signal when it falls below. This treats the SMA as a directional filter. It is structurally identical to a simple moving average strategy and does not use the bands themselves. The middle band crossover strategy generates signals only at the point of crossing: a long signal when the close moves from below to above the SMA, and a short signal for the inverse. This differs from trend following by requiring a fresh cross rather than sustained position above or below. Of these six, only band penetration and Percent B threshold have a connection to the volatility-relative mean reversion framework that Bollinger (2001) described. The band touch strategy misinterprets the bands as support and resistance levels. The squeeze strategy draws on Bollinger's bandwidth observations but adds a directional prediction he never endorsed. The trend and crossover strategies ignore the bands entirely and use only the middle band, which is just a simple moving average. 4. Data and methodology 4.1 Asset universe We tested 14 liquid ETFs across five asset categories. US equities comprised SPY, QQQ, IWM, and DIA. International equities included EFA, EEM, and VWO. Commodities covered GLD and SLV. Bonds included TLT. Sector ETFs spanned XLV, XLE, XLF, and XLK. All data is daily, sourced from TwelveData with Tiingo as fallback, covering approximately 5,000 trading days per asset, spanning roughly two decades. 4.2 Parameter grid Band periods range from 5 to 200 in steps of 1, giving 196 values. Standard deviation multipliers range from 0.5 to 4.0 in steps of 0.25, giving 15 values. Holding periods span 35 values from 1 to 252 trading days: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 18, 20, 22, 25, 28, 30, 35, 40, 45, 50, 60, 75, 90, 100, 120, 150, 180, 200, and 252. Percent B thresholds use 7 values from 0.0 to 0.30. Squeeze percentiles use 5 values from the 5th to the 25th percentile of rolling 252-day bandwidth. Band touch, band penetration, Percent B threshold, and squeeze depend on both band period and deviation multiplier, generating 196 times 15 equals 2,940 band configurations per asset. Trend following and crossover depend only on the band period, not the deviation multiplier, as they use the middle band exclusively. The total configuration count across 14 assets is 20,360,480 target tests. After filtering for sufficient data length and a minimum of 10 signals per configuration, 19,765,587 valid tests remain. 4.3 Forward return measurement Edge is measured as the difference between mean forward returns following a signal and mean forward returns across all bars in the same asset sample. This baseline adjustment ensures that strategies in rising markets do not receive credit for capturing beta. Forward returns are computed from the close of the signal bar to the close of the bar at the specified holding period distance. 4.4 Statistical framework Significance is assessed using Welch's t-test for unequal variances, comparing returns following the signal against baseline returns. Given 19,765,587 tests, the Bonferroni-corrected significance threshold is 2.53 times ten to the negative ninth power. A result surviving this threshold would occur by chance fewer than once in 400 million tries under the null hypothesis. 5. Results 5.1 Overview Figure 1 presents the aggregate picture. Band penetration produces the widest positive distribution on the long side, with the interquartile range visibly above zero. Percent B threshold shows a similar but narrower positive distribution. The remaining four strategies cluster symmetrically around zero, with squeeze and crossover compressed most tightly. Across all 19,765,587 tests, mean long edge is positive 0.29 percentage points and mean short edge is positive 0.17 percentage points. These averages are misleading because they merge strategies that generate strong positive edge with strategies that generate negative edge, partially canceling one another. The decomposition by strategy reveals the structure. 5.2 Results by strategy Band penetration is the strongest strategy. From 1,264,895 tests, mean long edge is positive 1.22 percentage points and mean short edge is positive 0.20 percentage points. 7,131 long and 24,156 short results survive Bonferroni correction. Nominal significance rates reach 35.1 percent on the long side and 34.7 percent on the short, roughly seven times the 5 percent rate expected from chance. The long edge of 1.22 percentage points is the highest mean strategy edge documented in this series. For context, VWAP mean reversion short signal edge was 0.89 percentage points, and MACD histogram divergence long edge was 0.32 percentage points. Band penetration long signals beat both. The mechanism is specific: when the close falls below the lower Bollinger Band, meaning price has reached an extreme relative to its own recent volatility, subsequent returns significantly exceed baseline. This is mean reversion from a volatility-adjusted extreme. The asymmetry between long and short is notable. Long edge from lower band penetration is 1.22 percentage points. Short edge from upper band penetration is 0.20 percentage points. Buying after extreme drops works roughly six times better than shorting after extreme rallies. This is consistent with the empirical observation that equity drawdowns are faster and sharper than rallies, creating more pronounced mean reversion opportunities on the downside. Percent B threshold extends the penetration signal across a range of thresholds. From 9,783,505 tests, mean long edge is positive 0.56 percentage points and mean short edge is positive 0.19 percentage points. 79,681 long and 207,579 short results survive Bonferroni correction, making this the largest single block of significant results at 287,260. The lower mean edge compared to band penetration reflects the inclusion of less extreme thresholds, which generate weaker signals. At threshold 0.0, the Percent B strategy is mathematically identical to band penetration. At threshold 0.30, it fires when price is merely in the lower 30 percent of the band range, a weaker condition. The Bonferroni count shows that even these less extreme signals contain information, but the strongest signals come from the most extreme deviations. Together, band penetration and Percent B threshold account for 318,547 of the 320,256 total Bonferroni-significant results: 99.5 percent. Mean reversion from Bollinger Band extremes is not one of several working strategies. It is the only working strategy. The band touch strategy, the "bounce off the lower band" technique taught across retail forums, produces negative long edge of minus 0.40 percentage points from 1,322,467 tests. Only 216 long results and 49 short results survive Bonferroni correction. These numbers are not zero, but the direction matters: the significant long results have negative edge, meaning the strategy significantly destroys value. When the low touches the lower band but the close recovers, the data says the recovery is temporary. The price tends to continue lower, making the touch a bearish signal, not the bullish one that retail interpretation assumes. The squeeze breakout, the strategy most associated with Bollinger Bands in popular trading culture, produces zero Bonferroni-significant results from 7,202,640 tests. Not one. Seven million attempts to find a parameter combination where narrow bandwidth followed by a middle band cross predicts direction. Zero successes. The mean long edge is negative 0.10 percentage points and the mean short edge is positive 0.15 percentage points, both within the noise range and both below transaction costs. Bollinger (2001) described the squeeze as a precursor to volatility expansion. He was correct about that: bandwidth does expand after periods of contraction. What the squeeze does not predict is the direction of the expansion. The bands compress, then they widen, and the subsequent move is equally likely to be up or down. The retail interpretation adds directional information that the indicator does not contain. The middle band crossover produces zero Bonferroni-significant results from 96,040 tests. This is structurally identical to a simple moving average crossover, and the result matches: crossing a moving average does not predict future returns. This finding is consistent with the MACD study, where line crossovers also produced zero significant results. Trend following based on the position relative to the middle band produces 286 long and 1,158 short Bonferroni-significant results from 96,040 tests, but with negative edge on both sides: minus 0.04 percentage points for long and minus 0.02 for short. The significant results are significant in the wrong direction. When price is above the SMA, subsequent returns are slightly worse than baseline. When price is below the SMA, subsequent returns are slightly better. This is, once again, mean reversion. Even the middle band, stripped of the standard deviation envelope, points toward reversion rather than trend continuation. 5.3 Statistical significance Figure 3 shows the p-value distribution. Under the null hypothesis of no predictive information, p-values distribute uniformly between zero and one. The observed distribution departs sharply from uniformity, with heavy concentration at low values. 25.1 percent of long signals and 26.9 percent of short signals achieve nominal significance at p less than 0.05, roughly five times the chance rate. From 19,765,587 tests, 4,957,687 long and 5,318,743 short results achieve nominal significance. After Bonferroni correction, 87,314 long and 232,942 short results survive. The 320,256 total exceeds the VWAP study's 150,546 and far exceeds MACD's 3,235. 5.4 Results by asset category US equities show the strongest effects: long edge positive 0.50 percentage points and short edge positive 0.40 percentage points. International equities follow with long edge positive 0.15 and short edge positive 0.37 percentage points. Sector ETFs show long edge positive 0.33 and short edge positive 0.11 percentage points. Commodities present an interesting asymmetry. Long edge is positive 0.24 percentage points, meaning buying after gold or silver drops below the lower band generates above-baseline returns. But short edge is negative 0.43 percentage points, meaning shorting after gold or silver rises above the upper band generates returns substantially worse than baseline. Commodity prices above the upper band predict continuation, not reversion. This is consistent with commodity price behavior during supply shocks and inflationary episodes, where momentum dominates over mean reversion. Bonds show minimal effects: long edge negative 0.16 and short edge near zero. The Bollinger Band framework adds little for treasury ETFs, likely because interest rate dynamics are driven by central bank policy rather than the supply-demand microstructure that generates mean reversion in equities. 5.5 Parameter sensitivity 03_bb_parameter_heatmap.jpeg Figure 5 maps the parameter landscape. The top-left panel shows long edge by band period and holding period. The strongest positive values concentrate at shorter band periods (5 to 40) combined with longer holding periods (60 to 252 days). Shorter band periods produce more volatile bands, generating more frequent extreme signals. Longer holding periods capture more of the mean reversion. The bottom-left panel shows long edge by deviation multiplier and holding period. The strongest effects appear at lower deviation multipliers (0.5 to 1.5) combined with longer holding periods. Lower multipliers produce narrower bands, meaning "below the lower band" requires a less extreme deviation. The fact that even moderate deviations (0.5 to 1.0 standard deviations) produce significant edge suggests the mean reversion effect is not limited to tail events. Figure 6 isolates the holding period effect. Long edge increases monotonically from near zero at 1-day holding to approximately 0.8 percentage points at 252-day holding. This mirrors the VWAP finding: mean reversion from extreme levels develops over weeks and months, not hours. The practical implication is favorable: longer holding periods reduce transaction cost impact and execution timing sensitivity. 6. Economic significance 6.1 Edge versus transaction costs The band penetration long edge of 1.22 percentage points is a gross figure before transaction costs. Round-trip costs for liquid ETFs, including spread, impact, and commissions, run approximately 0.10 to 0.15 percentage points. That leaves net edge of approximately 1.07 to 1.12 percentage points, a ratio of roughly 8:1 between gross edge and costs. By the standards of this series, which considered VWAP's 6:1 ratio noteworthy, the Bollinger Band penetration ratio is the highest we have documented. The Percent B threshold strategy, with mean long edge of 0.56 percentage points, provides a 4:1 ratio. This is lower but still comfortably above the 2:1 threshold generally considered minimum for practical implementation. 6.2 Signal frequency Extreme band violations are rare by definition. A close below the lower band at two standard deviations occurs on roughly 5 percent of trading days for a normally distributed series, and less frequently for actual equity returns, which exhibit negative skewness. For a 14-asset universe, this generates on the order of a few signals per month. The signal density is sufficient for systematic implementation but not for high-frequency trading. This constraint mirrors VWAP mean reversion: the edge exists because the signals are infrequent enough that arbitrage capital cannot concentrate sufficiently to eliminate the effect. 7. Why band penetration works and squeeze does not 7.1 The volatility-adjusted extreme DeBondt and Thaler (1985) documented that stocks experiencing extreme price declines over three to five year horizons tend to outperform in subsequent periods. Jegadeesh (1990) found similar reversal effects at shorter horizons. Poterba and Summers (1988) presented evidence of mean reversion in aggregate stock market returns. The mean reversion literature is extensive and consistent: extreme price movements tend to partially reverse. Bollinger Bands offer a specific implementation of this principle by defining "extreme" relative to the current volatility regime. When volatility is high, the bands are wide, and reaching the lower band requires a larger absolute price move. When volatility is low, the bands are narrow, and a smaller move suffices. This adaptive calibration means the signal fires when price is unusual relative to recent conditions, not relative to an arbitrary fixed threshold. RSI defines oversold as below 30 regardless of market conditions. Bollinger Bands define it relative to what has been happening. The data suggests this distinction matters. The mechanism is the combination of two well-documented tendencies: price mean reversion from extremes and volatility mean reversion from elevated levels. When the close falls below the lower band, price has dropped to a level that two standard deviations of recent volatility would not explain. Both the price level and the volatility level that generated the signal are likely to revert, and both reversions benefit the long position. The double reversion creates a compounding effect that a price-only measure does not capture. 7.2 Why the squeeze fails The squeeze, narrow bandwidth followed by a directional break, relies on a single correct premise and a single incorrect one. The correct premise is that low volatility precedes high volatility. Volatility clustering, as documented by Engle (1982), implies that periods of compression are followed by expansion. The incorrect premise is that the direction of the expansion is predictable. Nothing in volatility clustering theory predicts direction. The bands tell you that a move is coming. They do not tell you which way. Seven million tests confirm this. Long signals from squeeze breakouts generate negative 0.10 percentage points of edge. Short signals generate positive 0.15 percentage points. Both are economically negligible and statistically indistinguishable from zero after Bonferroni correction. The squeeze tells you something real about future volatility. It tells you nothing about future direction. Retail traders interpreted the first fact as implying the second. The data says otherwise. 7.3 The band touch illusion The interpretation of Bollinger Bands as dynamic support and resistance is conceptually appealing but empirically wrong. The band touch strategy shows negative long edge of minus 0.40 percentage points. When the low touches the lower band but the close recovers, the recovery is unreliable. Subsequent returns are below baseline. The touch is not support. It is the first sign of an impending band penetration. This makes sense on reflection. If the low has already reached the lower band within a single session but the close managed to recover, the selling pressure that pushed price to the band still exists. The intraday recovery does not resolve the condition that created the extreme. Band penetration, where the close itself breaks through, represents a more complete expression of the dislocation, and the mean reversion from that more extreme condition is what the data supports. 7.4 Reconciling with the price-only failure thesis The first four studies in this series established that price-only transformations fail. RSI and MACD, both derived exclusively from closing prices, produce nothing of value. VWAP, which multiplies price by volume, produces substantial edge. Bollinger Bands are technically price-only. They use closing prices and nothing else. Yet they produce 320,256 Bonferroni-significant results, more than VWAP's 150,546. Does this contradict the thesis? It refines it. The thesis was too narrow. The correct formulation is not that price-only indicators fail, but that first-order transformations of price fail. RSI computes a ratio of up moves to down moves. MACD computes differences between exponential moving averages. Both are operations on the level and direction of price changes. Neither accesses a property of price that the original series does not already contain. Standard deviation is a second-order statistic. It measures the dispersion of returns, which is volatility. Volatility is not encoded in the price level or the direction of price changes. It requires computing the variance of price changes over a window, a calculation that extracts genuine additional information about market state. The Bollinger Band lower boundary at SMA minus two standard deviations combines a first-order statistic (the average level) with a second-order statistic (the dispersion around that level). The combination defines a region in price space that reflects market conditions in a way that neither statistic alone could. This reconciliation extends the thesis rather than breaking it: indicators predict when they access information beyond the first-order price series. Volume provides one such source. Volatility provides another. Calendar effects provide a third. First-order transformations that merely repackage price level and direction provide none. 8. Comparison with previous studies Five indicators, 66 million tests, one framework. The record: RSI: zero Bonferroni-significant results from 26 million tests. A first-order nonlinear transformation of price that produces random output. Turn of the Month: 21 Bonferroni-significant results from 385 tests. A calendar effect driven by identifiable institutional flow cycles. Small test universe, real anomaly. VWAP: 150,546 Bonferroni-significant results from 5.8 million tests. Volume-weighted mean reversion with short edge of 0.89 percentage points. Mechanistically grounded in institutional execution benchmarking. MACD: 3,235 Bonferroni-significant results from 14.3 million tests. Histogram divergence with long edge of 0.32 percentage points. A faint signal at the boundary of detection. Bollinger Bands: 320,256 Bonferroni-significant results from 19.8 million tests. Band penetration long edge of 1.22 percentage points. Volatility-adjusted mean reversion. Two patterns now span 66 million tests. First, mean reversion from extreme levels is the consistent source of edge in technical analysis. VWAP mean reversion works. Bollinger Band penetration works. Both identify price at an unusual distance from a reference level and profit from the correction. The strategies that retail traders prefer, crossovers, momentum signals, breakouts, and trend following, fail consistently across every indicator tested. Second, the information source matters, but the constraint is more nuanced than "price-only fails." First-order transformations of price (RSI, MACD) fail. Second-order statistics (Bollinger's standard deviation) and additional data dimensions (VWAP's volume, the calendar's timing) succeed. The common thread is that the successful indicators access a property of the market that the price level alone does not reveal: volatility regime, volume distribution, or institutional flow cycle. 9. Implications For traders using Bollinger Bands as support and resistance: the band touch strategy produces negative long edge. Using the lower band as a buy zone generates returns below baseline. The strategy does not work and does not come close to working. The band is not support. It is a warning. For traders waiting for the Bollinger Squeeze: 7,202,640 tests. Zero Bonferroni-significant results. Bandwidth compression predicts volatility expansion. It does not predict direction. A squeeze entry strategy requires a separate edge to determine direction, and the data shows Bollinger Bands do not provide one. For systematic strategy developers: band penetration on the long side represents the strongest single-strategy finding in this five-study series. The 1.22 percentage point long edge at an 8:1 ratio to transaction costs is economically meaningful. The effect concentrates in US equities, strengthens with holding period, and works across a wide range of band periods and deviation multipliers, reducing the risk that the result depends on a narrow parameter sweet spot. The practical implementation path is a diversified equity ETF universe with band penetration signals, position sizing proportional to deviation magnitude, and holding periods measured in weeks to months. For users of the standard 20/2 setup: the parameter heatmaps show that the standard configuration falls within the profitable region but is not at the optimum. Shorter band periods and lower deviation multipliers generate more signals with slightly lower per-signal edge but higher aggregate opportunity. The standard 20/2 is a reasonable starting point, not a constraint. For trading educators: two strategies should be retired from Bollinger Band curricula. The squeeze breakout, tested across seven million configurations, produces nothing. The band touch as support and resistance produces negative results. What the data supports teaching is the mean reversion interpretation: when price violates the band, the violation tends to correct, and the correction is the edge. Bollinger himself described the bands as defining relative highs and lows. The relative low is the trade. 10. Limitations Several constraints bound these conclusions. First, the analysis uses daily data only. Bollinger Bands applied to intraday timeframes were not tested. The VWAP study showed that timeframe affects results substantially, and band penetration might behave differently at higher frequencies. Second, the study tests each strategy in isolation. Combinations of band penetration with volume filters, volatility regime detection, VWAP mean reversion, or the Turn of the Month effect could alter results. Third, execution is assumed at the close of the signal bar. Band penetration signals fire at market close, meaning the practical entry occurs at the next day's open. Overnight gaps could reduce or augment the observed edge. Fourth, the band touch and band penetration strategies are defined using daily OHLC data. The intrabar path matters for the touch strategy, and different data sources may produce different signals for the same trading day. Fifth, position sizing was not modeled. Scaling position size with the magnitude of band violation, entering larger positions when deviation is extreme, could substantially improve risk-adjusted returns given the nonlinear relationship between deviation size and subsequent reversion. Sixth, transaction cost estimates reflect current market conditions for liquid ETFs. Less liquid instruments or historical periods with wider spreads would face higher cost drag against the observed edge. 11. Conclusion 19,765,587 parameter configurations. Six strategy types. Fourteen assets. Five categories. 320,256 Bonferroni-significant results. John Bollinger designed the bands as a relative framework for defining high and low prices. He did not claim they were a trading system. This analysis shows that they contain predictive information, but the prediction is narrow. Price that closes below the lower Bollinger Band, an event that means the close has reached an extreme relative to its own recent volatility regime, tends to revert. The long signal edge of 1.22 percentage points for band penetration is the strongest single-strategy result in five studies spanning 66 million tests. The analysis is equally clear about what does not work. The Bollinger Squeeze produces zero significant results from over seven million tests. The middle band crossover produces zero. The band touch strategy, the "bounce off the band" technique, produces significant results in the wrong direction. Four of six strategies fail. The finding fits the larger pattern emerging from this series. Mean reversion from extreme levels, whether measured by VWAP deviation or Bollinger Band penetration, is the consistent source of edge in systematic technical analysis. The popular strategies, crossovers, breakouts, squeezes, and support and resistance interpretations, fail consistently across every indicator tested. Sixty-six million tests, five indicators, and the conclusion is the same each time: the market reverts from extremes, and everything else is noise. Bollinger's original insight was that high and low should be defined relative to volatility. The twenty million tests confirm that he was right about the framework. The retail community was wrong about the application. References Bollinger, J. (2001). Bollinger on Bollinger Bands. McGraw-Hill, New York. DeBondt, W.F.M. and Thaler, R.H. (1985). Does the stock market overreact? Journal of Finance, 40(3), pp. 793-805. Engle, R.F. (1982). Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation. Econometrica, 50(4), pp. 987-1007. Fama, E.F. (1970). Efficient capital markets: A review of theory and empirical work. Journal of Finance, 25(2), pp. 383-417. Jegadeesh, N. (1990). Evidence of predictable behavior of security returns. Journal of Finance, 45(3), pp. 881-898. Lo, A.W., Mamaysky, H. and Wang, J. (2000). Foundations of technical analysis: Computational algorithms, statistical inference, and empirical implementation. Journal of Finance, 55(4), pp. 1705-1765. Park, C.H. and Irwin, S.H. (2007). What do we know about the profitability of technical analysis? Journal of Economic Surveys, 21(4), pp. 786-826. Poterba, J.M. and Summers, L.H. (1988). Mean reversion in stock prices: Evidence and implications. Journal of Financial Economics, 22(1), pp. 27-59. Sullivan, R., Timmermann, A. and White, H. (1999). Data-snooping, technical trading rule performance, and the bootstrap. Journal of Finance, 54(5), pp. 1647-1691.