Research · Validation
Technical Signals, Measured
The technical indicators traders commonly watch, tested rather than assumed to work. RSI, moving averages, MACD and volatility compression are shown live and evaluated against historical baselines. A signal matters here only if outcomes after the trigger were meaningfully different from outcomes after comparable random days.
Current readings
Computed from the daily public snapshot for 14 October 2026, published 2026-10-15 11:22 UTC. The charts and tables below are drawn in the browser from the same file; these figures are the ones they show.
| Price vs 200-day average | 15.2% | 200-day $72,129 |
|---|---|---|
| Realised volatility, 30 day | 39.4% | DVOL 39.1 |
| Variance risk premium | -0.2 points | 27th percentile |
TODAY'S READINGS
Buying on a randomly chosen day since 2017 was followed by a median gain of about 48% over the next year. Measured over the full history since 2010 the median is about 94%, but that figure sits on a very wide spread (-51% at the tenth percentile, +938% at the ninetieth, with 28% of days followed by a loss) and roughly half of it comes from 2010 to 2012. Every difference in the table is measured against an eligibility-matched baseline: the median over the days on which that rule could have fired (its inputs existed), from 4,787 to 5,536 days at one year depending on the rule, so a rule is never compared with days it could not see. The same baseline sits on both sides of every interval. That is the number every signal here has to beat, and once the overlap between windows is accounted for, 6 of the 8 rules tested show no clear evidence of beating or trailing it, while one shows evidence of trailing it, and 1 shows evidence of an edge. One rule does stand apart, and it is the opposite of what it is taught to mean: after relative strength rose above 70 — the “overbought” level usually read as a warning — returns ran ahead of the baseline, and the interval stays above zero at one month and three months. At the horizons not listed the interval includes zero, so the finding is clearest in the near term rather than over a year.
A rule counts as evidence only where its confidence interval excludes zero at some horizon under three independent bootstrap seeds; differences under 5 points are merely coloured neutrally, which is a reading convention and not a test. Nor is any of this a recommendation. This page cannot tell you to buy, hold or sell: that depends on your horizon, your position size and whether a deep fall would force you out, none of which are in this data. What it can tell you is that the readings below are weaker evidence than they are usually presented to be — and today they disagree with each other, with relative strength stretched upward while the 50-day average sits below the 200-day.
Today's readings, by signal type
| Family | What it answers | Reading now | State |
|---|---|---|---|
| Momentum | Is the move stretched, and in which direction? | Relative strength 51.9, MACD histogram -526 | neither |
| Trend | Has a change already been confirmed? | 50-day $81,105 against 200-day $72,129, price above the 200-day | 50-day above the 200-day |
| Volatility | Is risk expanding or compressing? | Bollinger width 5.8% of its middle band, realised volatility 39% annualised | compressed |
| Participation | Is capital confirming the move? | measured on the Flows and Positioning page, not here | see Flows |
Participation and positioning — open interest, funding, options skew, ETF flows and CME positioning — are deliberately not repeated here; they are measured on the Flows and Positioning page. Exchange volume, cumulative delta and liquidations are absent from the whole site because the snapshot has no free, keyless, primary source for them; on-chain transfer value is not a substitute and is not presented as one.
What each signal has been worth
| Signal | Triggers | 1 month | 3 months | 1 year | Verdict |
|---|---|---|---|---|---|
| EARLY WARNING · Can move before a trend turns, and fires often when nothing follows | |||||
| Relative strength rises above 70, the textbook overbought line | 38 | +10.4 +5 to +16 * | +25.8 +1 to +37 * | +51.4 -47 to +181 | evidence of an edge |
| Relative strength falls below 30, the textbook oversold line | 29 | -2.9 -8 to +6 | +5.8 -25 to +23 | -17.2 -127 to +136 | no clear evidence |
| Bollinger width enters the bottom tenth of its trailing two years | 26 | -4.6 -14 to -1 * | -2.8 -15 to +12 | -10.7 -61 to +84 | evidence against |
| Realised volatility enters the bottom tenth of its trailing two years | 27 | -5.5 -12 to +2 | -7.2 -17 to +20 | +23.6 -39 to +108 | no clear evidence |
| CONFIRMING · Only fires after the move that produced it | |||||
| Golden cross, 50-day rises above 200-day | 13 | +0.3 | +0.6 | +13.6 | no clear evidence |
| Death cross, 50-day falls below 200-day | 13 | -5.3 | +15.5 | -71.3 | no clear evidence |
| MACD rises above its signal line | 115 | +0.8 -3 to +3 | +2.2 -17 to +12 | -15.9 -121 to +56 | no clear evidence |
| MACD falls below its signal line | 113 | +0.4 -3 to +3 | -2.9 -23 to +4 | -14.6 -134 to +34 | no clear evidence |
Each cell is the difference, in percentage points, between the return after the signal and the return after a randomly chosen day. Beneath each is a ninety per cent bootstrap interval for that difference, a robustness estimate under the stated dependence structure rather than a model-based probability, obtained by resampling both sides in blocks — the baseline as contiguous calendar days, the triggers as blocks of the ordered event sequence so that neighbouring overlapping episodes stay together — marked with an asterisk where it excludes zero under all three seeds. A verdict of evidence requires that mark; the five-point rule only colours the figure. Full returns are in the table below.
Show the underlying returns for every signal and horizon
| Signal | Horizon | After the signal | After any eligible day | Historical difference | Times it rose |
|---|---|---|---|---|---|
| Relative strength rises above 70, the textbook overbought line | 30d | +13.6% | +3.2% | +10.4 pts | 74% |
| ⋮ | 90d | +37.8% | +12.0% | +25.8 pts | 76% |
| ⋮ | 365d | +144.6% | +93.3% | +51.4 pts | 76% |
| Relative strength falls below 30, the textbook oversold line | 30d | +0.3% | +3.2% | -2.9 pts | 55% |
| ⋮ | 90d | +17.8% | +12.0% | +5.8 pts | 62% |
| ⋮ | 365d | +76.0% | +93.3% | -17.2 pts | 70% |
| Bollinger width enters the bottom tenth of its trailing two years | 30d | -1.6% | +3.0% | -4.6 pts | 46% |
| ⋮ | 90d | +6.3% | +9.1% | -2.8 pts | 56% |
| ⋮ | 365d | +65.2% | +75.9% | -10.7 pts | 74% |
| Realised volatility enters the bottom tenth of its trailing two years | 30d | -2.5% | +3.0% | -5.5 pts | 41% |
| ⋮ | 90d | +2.0% | +9.1% | -7.2 pts | 54% |
| ⋮ | 365d | +99.5% | +75.9% | +23.6 pts | 79% |
| Golden cross, 50-day rises above 200-day | 30d | +3.3% | +3.0% | +0.3 pts | 54% |
| ⋮ | 90d | +10.7% | +10.1% | +0.6 pts | 67% |
| ⋮ | 365d | +100.7% | +87.1% | +13.6 pts | 75% |
| Death cross, 50-day falls below 200-day | 30d | -2.3% | +3.0% | -5.3 pts | 46% |
| ⋮ | 90d | +25.6% | +10.1% | +15.5 pts | 54% |
| ⋮ | 365d | +15.8% | +87.1% | -71.3 pts | 50% |
| MACD rises above its signal line | 30d | +4.0% | +3.1% | +0.8 pts | 59% |
| ⋮ | 90d | +14.4% | +12.2% | +2.2 pts | 59% |
| ⋮ | 365d | +77.8% | +93.7% | -15.9 pts | 69% |
| MACD falls below its signal line | 30d | +3.5% | +3.1% | +0.4 pts | 58% |
| ⋮ | 90d | +9.3% | +12.2% | -2.9 pts | 57% |
| ⋮ | 365d | +79.1% | +93.7% | -14.6 pts | 69% |
- The textbook is upside downbuying when relative strength crossed above 70, the level that is supposed to mean overbought, beat a randomly chosen day by +51.4 pts at one year; crossing below 30, the supposed bargain, came in -17.2 pts at the same horizon
- The MACD found nothingits two crossings differ from a random day by +0.8 pts and +0.4 pts at thirty days, which on this sample is indistinguishable from no information at all
- The crosses are latethe golden cross came in +0.3 pts at thirty days: by the time a 50-day average clears a 200-day one, the move that lifted it has already happened
- Compression promises nothinga Bollinger-width squeeze into the bottom tenth of its own two-year range came in -4.6 pts at thirty days, and a realised-volatility squeeze -5.5 pts; the coiled-spring reading does not say which way the spring goes
- Read the direction, not the decimaltrigger counts run from twelve to over a hundred, the windows overlap, and the baseline itself is volatile; treat a few points of edge as noise
Price, the two moving averages, and relative strength
Why there are no Fibonacci levels here
Fibonacci retracements depend on which high and low the analyst chooses. Change those points and every level changes. Without a precise rule for choosing the swing points, the method cannot be tested consistently. The same principle applies to trend lines, channels and many chart patterns: if a rule can be specified precisely enough to program, it can be tested: there is no rule this page could code, run over the history, and report a hit rate for. The same objection applies to trend lines, channels and most chart patterns. They are omitted not because they are unpopular but because nothing here can be measured, and this site does not publish what it cannot measure. If a reader can state such a rule precisely enough to program — which swing points, over what window, with what tolerance — it can be added to this table and tested like the rest.
Method
All series are computed in the browser from the daily closes in the snapshot. Relative strength uses Wilder's smoothing over fourteen days; the moving averages are simple; the MACD is the twelve and twenty-six day exponential averages with a nine-day signal line, as originally specified. A trigger is the day a condition first becomes true, and repeat triggers within ninety days are counted once for the moving-average and relative-strength rules and within thirty days for the MACD, which fires far more often. Forward returns are simple price changes.
The baseline column is the median return over the same horizon measured from every day on which the rule could have fired: the days on which its inputs existed, with a completed forward window. It is the return a reader would have got by choosing one of those days at random, and it is high because Bitcoin rose enormously over the period. That median describes a badly skewed distribution rather than a typical outcome, and it is sensitive to where the history starts: roughly half of it comes from 2010 to 2012, when Bitcoin moved from cents to dollars, and measuring only from 2017 lowers it substantially. The full history is used because it sets the harder benchmark for any signal to beat, and every comparison applies the same baseline to both sides. The baseline is eligibility-matched: the moving-average rules cannot fire before day 200 and the squeeze rules before day 750, so their baselines exclude those days rather than compare a rule with days it could not see. The unmatched full-history median is shown in the banner for context only. Any signal must beat it to have said anything at all, and this is where most published back-tests quietly fail: they report the return after the signal without reporting the return without it.
What this page tells you, and what it does not
The page shows: these are the historical records of each rule on this data, reproducible from the snapshot. It does not claim: that any of them will behave the same way again, or that the edges shown are statistically significant. Trigger counts here run from twelve to a hundred and fourteen as of 14 October 2026 and the windows overlap heavily: every consecutive pair of triggers in this sample has overlapping one-year windows. The intervals shown resample both sides of the comparison: the baseline is the daily series of forward returns and is drawn as contiguous calendar blocks, which is a moving-block bootstrap in the strict sense, while the trigger side is drawn as blocks of the ordered event sequence, which is not one and is not called one. Resampling both means the interval describes uncertainty in the difference of two estimated medians rather than in the signal median alone. Where a signal has fewer than twenty completed windows no interval is computed at all. Read the direction and the size, not the decimal.
Sources: the Crypto Exponentials daily snapshot (Blockchain.com daily close). Signal definitions: J. Welles Wilder, New Concepts in Technical Trading Systems, 1978, for relative strength; Gerald Appel for the MACD. Nothing here is investment advice.