Research
Replications: do published findings hold on crypto?
Each note takes a well-known claim — from a paper, a strategy library or market commentary — tests it on our own crypto market data, mostly bitcoin and ether, and gives the verdict in the first paragraph. The original source is cited first; the data, the test and the verdict are ours. Most claims do not survive, and each note says what would change that.
Verdicts: Replicated — the claim holds on our data; Partly replicated — part of the claim holds, part does not; Did not replicate — the claim does not hold on our data; Inconclusive — our data cannot settle it either way.
- Can a GARCH model predict bitcoin’s direction 85% of the time?
No. Rebuilt from a 2023 paper on bitcoin perpetual futures, the EGARCH-based trading model called the next bar’s direction correctly 50.1–52.2% of the time from 2020 to October 2026, at horizons from 15 minutes to a day — no better than the sign of the recent average return — and after costs it trailed buy-and-hold. We could reproduce an 85% figure only by scoring the model against the window it was fitted on.
- Do token unlocks push crypto prices down?
Not by much, as far as our data can tell. Across vesting tokens with perpetual futures from 2021 to September 2026, prices fell about 2% more than a placebo in the 30 days before unlocks of at least 1% of supply — a third of the size reported in industry studies and not distinguishable from noise. Large unlocks of 5% or more looked worse, about −6%, but not reliably.
- Is there a day-of-the-week effect in bitcoin?
Not in our data. Between August 2017 and August 2026 bitcoin averaged +0.33% on Mondays and +0.35% on Wednesdays against −0.34% on Thursdays, but in no market era is the spread between days larger than the noise you get when seven weekdays are compared at once. Rolling weekday averages, as shown on dashboards, had no forecasting value.
- Do moon phases affect bitcoin’s price?
No. Being long bitcoin in the half of the lunar month around the new moon and short around the full moon — the rule behind the “lunar effect” in stocks — produced a Sharpe ratio of −0.04 on bitcoin and −0.14 on ether from August 2017 to September 2026. The new-moon half returned slightly less than the full-moon half, the opposite of the claim.
- Does bitcoin follow gold with a lag?
We found no lead. Testing gold’s moves against bitcoin and ether at daily, weekly and monthly lags from August 2017 to September 2026, no lag predicted crypto returns once the number of lags searched is taken into account. The two do move together on the same day, but weakly: the daily correlation is about +0.12, so gold explains roughly 1.5% of bitcoin’s daily variance.
- Is bitcoin’s mining cost a price floor?
Not as a trading signal. From January 2015 to August 2026, rules that held bitcoin when its price was cheap relative to the estimated cost of mining reached a Sharpe ratio of 0.76 at best, below a rule that knows only the number of days since the last halving (1.07). Each halving doubles the cost overnight, so “cheap versus cost” mostly means “shortly after a halving”.
- Can machine learning predict bitcoin’s daily direction?
Not in our replication. A calibrated XGBoost classifier built to the recipe of a 2026 paper called bitcoin’s next-day direction correctly on 49.8% of days from 2022 to 2026, and its probability forecasts scored worse than a simple base rate in every year. The paper’s surviving configurations all decide at 21:30 UTC, where, in the authors’ public code, some inputs already include part of the day being predicted.
- Do candlestick patterns predict crypto prices?
Not after costs, in our test. Of the seven candlestick patterns a 2026 study singled out as predictive on hourly crypto data, only two kept their sign on 2022–2026 perpetual futures, at a fraction of the reported size, and both effects disappear when the trade is entered one minute after the candle closes. No pattern covered a typical altcoin trading cost.
- Does TD Sequential work on bitcoin?
Not on hourly bars. Buying bitcoin and ether perpetual futures on TD Sequential’s “buy 9” setup, as coded in a public strategy library for the freqtrade bot, lost money before any costs from 2020 to September 2026 (gross Sharpe ratio −0.75 for bitcoin, −0.20 for ether), while the same exposure at random times earned +0.39 and +0.44. After nine lower closes, prices tended to keep falling.
- Can you trade bitcoin’s reaction to CPI and jobs reports?
The reaction is real but is over within a minute or two. From December 2021 to September 2026 bitcoin fell after hotter-than-expected CPI and payroll numbers and rose after cooler ones, as a 2026 study reports, but entering two minutes after the release and holding for 15 to 60 minutes did not earn more than its costs, and since 2024 the move after entry has been close to zero.
- Can you collect perpetual funding by entering just before it is paid?
No. On BTC, ETH and ten liquid altcoin perpetual futures from January 2022 to September 2026, the price moved against the side receiving funding by about as much as the payment — roughly 1.2 basis points of adverse move per basis point of funding — in the minutes around each 8-hour settlement. Gross profit was near zero at every funding level, and after fees the trade lost 0.03% to 0.12% per round trip.
- Can an AI foundation model predict crypto prices?
Not the direction, in our test. Kronos, an open foundation model pre-trained on 12 billion candlesticks, called the sign of bitcoin’s and ether’s next 1, 4 and 24 hours correctly 49.8% to 51.6% of the time on data after its training cut-off (July 2024 to October 2026). The best configuration looked good before costs but did not stand out from chance, and the hourly versions lost money after fees.
- Does the turn-of-the-month effect work in bitcoin?
Not reliably, and not since 2024. From September 2017 to October 2026 bitcoin and ether did better on the last day and first three days of each month than in other four-day windows (+0.19% and +0.35% a day), but the premium came entirely from the years before 2024 and has been negative since the spot ETF launch. Holding only at the turn of the month lost to buy-and-hold in 2024–2026.
- Does bitcoin rise in options-expiration week?
No — if anything it has done worse. From September 2017 to September 2026, bitcoin returned 0.19% a day less, and ether 0.37% a day less, in the week leading up to the monthly options expiry than in the same hours of other weeks. The stock-market finding that expiration weeks earn higher returns does not carry over to crypto, though the negative gap is too noisy to rely on either.
- Do liquidity sweeps predict bitcoin’s next move?
Not beyond chance, in our data. Running a published test protocol on bitcoin from January 2020 to October 2026, the gap between “accepted” and “rejected” sweeps of monthly, weekly and round-number levels was the same at real levels as at random price points — 0.117 versus 0.114 of a typical daily range over the next 24 hours. The level itself added nothing.
- Can the Hurst exponent tell when bitcoin is trending?
No. On rolling 60-day windows the standard rescaled-range estimate of the Hurst exponent reads about 0.59 even on shuffled bitcoin returns, which have no memory, so “H above 0.5” flags trends that are not there. Used as a switch between momentum and mean reversion on bitcoin and ether from 2017 to 2026, it did worse than plain momentum, and high readings preceded weaker momentum, not stronger.
- Does raising leverage as price falls improve leveraged DCA?
Not beyond arithmetic. Raising leverage at each step of a dip-buying ladder lowers the average entry price by construction, but on twelve crypto perpetual futures from 2020 to 2026 its advantage over a decreasing schedule (+0.44 points of budget per episode) was smaller than the same ladders show on directionless random paths (+0.76), and it did no better than equal leverage. Liquidation rates were set by the ladder’s geometry.
Research, not investment advice.