Buy the Dip or Catch a Falling Knife? Test the Confirmation

Buy the dip or catch a falling knife? The empirical answer

A 10% rolling drawdown alone did not distinguish a bargain from a falling knife in these 231 complete Binance BTC/ETH/SOL/BNB/XRP/DOGE spot opportunities, valued at the shared signal+31-day open endpoint after 0.20% per-side allowances. Immediate entry lost in 123/231 (53.2%) cases. Its positive mean of +5.91% and negative median of -1.01% describe a skewed distribution: some large rebounds lifted the average while more than half the observations still lost. A mean alone would obscure the typical experience.

The combined confirmation policy entered 84 times. It had 45 profitable trades, 39 losing trades and 147 no-entry cash outcomes. Its 10th-percentile opportunity return was -14.01% versus -23.57% for immediate entry, but its mean paired difference was -3.51 pp with a 95% quarter-clustered interval of -11.15 to +1.72 pp. That interval crosses zero; an average-return advantage is not established.

The actionable distinction is not “a red candle means buy.” Separate the size of the drop, the intensity of selling-related activity and whether a later close actually reclaimed a short-term high. Waiting changes participation and loss exposure; it also misses rebounds. This is an entry-policy experiment, not a claim to identify the final market bottom.

Exactly what counts as a dip, a bounce and confirmation?

A candidate appears when a daily close first crosses to 10% or more below the highest close in the previous 20 UTC days. Yesterday must still have been above its own -10% threshold. This is a rolling-close drawdown, not a one-day 10% crash, all-time-high drawdown or an intraday wick.

We require 200 prior days and impose a 30-day cooldown per symbol. The same declining episode therefore cannot generate a new buy every day. The six markets are BTC, ETH, SOL, BNB, XRP and DOGE against USDT on Binance; choosing current survivors creates survivorship bias. Missing candles are not imputed.

Immediate entry buys next day’s open. Confirmation waits at most seven subsequent closed days for a close above the highest high of the preceding three days. Combined confirmation also requires that day’s valid taker-buy quote volume / total quote volume be at least 50%. Both enter at the following open. All sell at signal date +31 days open. No stop loss or leverage is modelled.

Cash that never receives confirmation stays at zero return and remains in the denominator. A zero median for a waiting policy is therefore not proof that its actual trades broke even. Reporting only entered trades would select a different, future-conditioned cohort.

Which indicators helped distinguish risky dips in this sample?

High-volume dips (at least 2× the preceding 20-day median quote turnover) had 13/42 (31.0%) profitable immediate entries and a -3.87% median, versus 95/189 (50.3%) and +0.37% below 2×. That is a useful warning to investigate, not evidence that volume caused the subsequent loss.

Neither being above SMA200 nor combining an above-SMA200 trend with active buying produced a sufficiently convincing universal buy rule. The combined trend-and-buying subset has only nine observations; the RSI≤30 subset only seven. They remain exploratory. We do not rename a small favourable subset a validated strategy.

These groups use information from the signal close. They are not matched for regime, symbol or volatility, and multiple comparisons are not adjusted. A future confirmation uses information that arrives later, so its entry is later too. Mixing those two clocks would create look-ahead bias.

Technical definitions: RSI, SMA, volume and active buying

SMA200 is the arithmetic mean of 200 daily closes including the signal close. Wilder RSI(14) initializes average positive and negative close changes over fourteen days, then smooths each using (13 × prior average + new change) /14. RSI =100−100/(1+average gain/average loss); a flat series is 50, zero losses with positive gains is 100.

Volume ratio uses today’s USDT quote turnover divided by the median of the previous twenty days, excluding today. Active-buy share is taker-buy quote turnover divided by all quote turnover, only when valid=1. It describes which side crossed the spread; it is not the number of buyers, exchange deposits or total capital entering crypto.

The confirmation threshold uses preceding daily highs, not only closes. A candle must close above that threshold. A brief wick or an unfinished candle does not qualify. There is no optimized RSI, volume or moving-average parameter chosen after seeing the results.

Is Bitcoin different from the other five coins?

The BTC subset has 32 complete opportunities and 14 positive immediate outcomes; its result cannot be replaced by the six-coin pooled rate. The table uses the same Binance USDT spot drawdown definition, next-open entry, shared signal+31-day exit and costs for each coin. The same rules are applied without coin-specific tuning. Per-coin results expose concentration and avoid allowing volatile altcoin rebounds to stand in for a BTC result. These are selected surviving liquid assets, not the entire Binance directory.

The analysis covers data from September 2020, with signal eligibility only after a full 200-day warmup. Recent opportunities without a complete 30-day outcome remain pending in the ledger, rather than being assigned zero.

Do trading costs and uncertainty erase the apparent advantage?

Waiting reduced the observed downside tail, but the combined-policy paired mean interval of -11.15 to +1.72 percentage points crosses zero and does not establish higher average returns in these 231 opportunities. The primary calculation charges 0.20% on purchase and sale. Sensitivities preserve the candidate cohort and entry times and use 0.10% or 0.50% per side. The next daily open is an executable-price proxy, not a promised fill; fast drops may incur more slippage.

For each waiting policy we compare its result with immediate entry on the exact same opportunity. All observations in a UTC quarter are resampled together, including different assets, for 1,000 draws with replacement using Python Random seed 18092026. Linear 2.5/97.5 percentiles define the interval; at least twenty observations and eight quarters are required. The interval estimates a paired mean difference, not a win-rate band, and cross-quarter dependence can remain.

Historical path: SOLUSDT · 2024-05-24

This Binance SOLUSDT · 2024-05-24 daily dip case uses a shared signal+31-day open endpoint and 0.20% per-side allowances. At the signal, RSI(14) was 54.9, quote-volume ratio 1.25× and active-buy share 53.8%. Immediate entry returned -23.57%. Price confirmation returned +0.00%; price plus active buys returned +0.00%. The combined policy did not enter; its zero is cash, not a winning trade.

Top: price and the 200-day simple moving average, indexed to the signal close. Middle: daily active-buy share, not deposits or exchange net flows. Bottom: the three policy account values. S = signal candle; I = immediate buy; B = price-confirmed buy; C = combined-confirmation buy; X = common exit. Absent entry markers mean no trade. Close values precede the final open endpoint; daily lines hide intraday risk.

Historical path: XRPUSDT · 2023-02-12

This Binance XRPUSDT · 2023-02-12 daily dip case uses a shared signal+31-day open endpoint and 0.20% per-side allowances. At the signal, RSI(14) was 40.9, quote-volume ratio 0.42× and active-buy share 45.9%. Immediate entry returned -1.01%. Price confirmation returned -7.50%; price plus active buys returned +0.00%. The combined policy did not enter; its zero is cash, not a winning trade.

Top: price and the 200-day simple moving average, indexed to the signal close. Middle: daily active-buy share, not deposits or exchange net flows. Bottom: the three policy account values. S = signal candle; I = immediate buy; B = price-confirmed buy; C = combined-confirmation buy; X = common exit. Absent entry markers mean no trade. Close values precede the final open endpoint; daily lines hide intraday risk.

Historical path: ETHUSDT · 2022-12-16

This Binance ETHUSDT · 2022-12-16 daily dip case uses a shared signal+31-day open endpoint and 0.20% per-side allowances. At the signal, RSI(14) was 39.6, quote-volume ratio 1.83× and active-buy share 48.3%. Immediate entry returned +32.52%. Price confirmation returned +27.07%; price plus active buys returned +27.07%. The combined policy did enter; its final result includes both execution-cost allowances.

Top: price and the 200-day simple moving average, indexed to the signal close. Middle: daily active-buy share, not deposits or exchange net flows. Bottom: the three policy account values. S = signal candle; I = immediate buy; B = price-confirmed buy; C = combined-confirmation buy; X = common exit. Absent entry markers mean no trade. Close values precede the final open endpoint; daily lines hide intraday risk.

A practical checklist for interpreting a falling coin

First define the drawdown reference: a fall from a twenty-day close high is not the same as a 24-hour ticker loss. Next check whether turnover is ordinary or exceptional; this study’s heavy-turnover group performed worse, despite the common intuition that all large-volume dips are accumulation.

Then separate a bounce from a completed reclaim. The tested confirmation needs a closed break above three prior highs; active-buy dominance adds context but is not independent proof of new capital. If neither condition occurs within the test window, this policy does nothing. Its missed trades are part of the cost of waiting.

Finally distinguish entry timing from risk sizing. No historical filter here establishes that a bottom cannot break. Our separate bottom-confirmation research asks whether subsequent lows hold; this article asks whether delaying a specific entry improves same-endpoint outcomes. Those are related but different questions.

A coin is down 10% today—what would make this a dip rather than a falling knife?

A coin down 10% over the latest 24 hours is not necessarily an event in this study: the tested trigger is a completed daily close first crossing 10% below its prior twenty-day closing high. For those 231 six-pair Binance spot opportunities, immediate entry had 108/231 positive signal+31-day outcomes and a -1.01% median net result. Combined price/active-buy confirmation kept 147/231 in cash and reduced the downside tail without establishing higher average returns. Wait is an explicit policy result, not missing data.

To assess a current coin, keep the timeframes and meanings separate: a rebound must be visible on completed price bars, valid buyer-initiated flow must be measured rather than assumed from turnover, and market-wide breadth should show whether the decline is isolated. RSI oversold alone is not a bottom; heavy volume during a fall may be urgent selling as well as later accumulation.

This research does not set your entry, stop or position size. A small newly listed token, a coin with an imminent unlock, or a leveraged futures position can have different risks from the six tested Binance spot pairs; do not transfer the sample's percentages to those cases.

Verify: Bitcoin bottom-confirmation research · Current market breadth and spot pressure

Reproduce the experiment: inputs, candidate ledger and field definitions

Download candles.json, derivatives.json and three_decision_studies.py into the same directory. Run python three_decision_studies.py . reproduced.json. The standard-library-only program rebuilds all three studies. Compare the result with research.json. The JSON keeps full precision; CSV nested objects are JSON-encoded, not silently truncated. Chart coordinates and selection rules are in chart-data.json.

The SHA-256 manifest verifies each published artifact. These are frozen September 18, 2026 results, not a live ticker. An update must publish a separate evidence edition and revise the answer, tables, FAQ and Dataset together. These public study files are free to download. Separately purchased fixed-edition data packs and optional API access have their own listed contents and permissions; this experiment does not expand those rights.

Binance candles were retrieved retrospectively. The input file records actual request URLs and retrieval timestamps; a historical candle date is not a CoinNudge first-observation timestamp. Every signal uses closed daily data, and execution uses a subsequent open. This reduces look-ahead within the calculation but does not turn retrospectively retrieved data into an original point-in-time archive.

Verify: Full analysis JSON · This study: complete candidate CSV · This study: summary CSV · Candle inputs JSON · Candle inputs CSV · Derivative input JSON · Derivative inputs CSV · Standalone reproduction program · Chart coordinates · Preset protocol · Integrity manifest

Frequently asked questions

Sources and verification

• Binance spot candlestick definitionsOfficial OHLC, quote-volume and taker-buy field definitions.
• Investor.gov: dollar-cost averagingDefinition of regular equal-sized investing, not evidence for this BTC result.
• Hyperliquid funding documentationHourly funding mechanics and payer direction; quotes are not guaranteed future settlements.
• CoinNudge experiment and evidenceFull-precision observations, denominators, groups and sensitivity.

Inspect the service before choosing a subscription.

See supported Telegram alerts and research-data samples. These experiments do not execute trades.