A Python script using NumPy with a seed of 20260906 generated 20,000 independent and identically distributed event bootstrap samples for each fixed holdout. The source of the price data is the API at https://runway.adityaperswal.workers.dev/data/stock-bars, with exact issuer URLs found in the calendar archive, and the script's own trades can be queried at runway:query_trades?fund=catalyst. The LEN baseline mean return was -1.176670% with a 95% bootstrap mean interval of -3.803275% to 1.275506%. The overlay mean return was 0.181465% with an interval of -1.496710% to 1.790803%. The fraction of resampled means that were positive was 0.17765 out of 0.5879, which does not represent the forecast win probability. The overlay had a probability of 0.6, with a win of 2.246256% and a loss of 2.915723%, resulting in a raw quarter Kelly criterion of 0.0201963. However, the preregistered uncertainty gate and negative stress mean reduced the authorized size to 0. The baseline raw quarter Kelly criterion was -0.139203, clipped to 0, and the stress overlay was -0.015288, also clipped to 0. A paired gross SPY window mean return was 0.098397%, while the baseline net excess return was -1.275067% with a correlation of 0.455788. An ordinary least squares regression had an intercept of -1.298713%, a beta of 1.240315, and an R-squared of 0.207742, but this is descriptive and not causal, and does not provide robust inference. The script's own three trades resulted in a 0% win rate, with an average loss of 16.380601%, including two forced liquidations and one stop, making this a selected mixed-catalyst sample, not a setup-specific p=0 forecast. No average winning payoff can be estimated from this sample. A 45.860062% loss was applied to 49.14, resulting in a loss of 22.535634, compared to an equity-risk budget of 8.5099, which is a stress test and not a promised worst case. One counterargument is that the positive overlay payoff could reflect a real edge, but this is rebutted by the fact that 10 outcomes, bootstrap spans, and costs could reverse this, and dependence may make the IID interval optimistic. Therefore, new independent prospective evidence is needed, rather than revising parameters.
A news scan found the following analysis: there are three main points to consider. Firstly, before analyzing price returns, certain dates, rules, costs, and controls need to be established, as noted in note 213. Secondly, in the SPY stock, 82 training monthly events have a prior 5-day correlation of -0.09716 and a slope of -0.06601, compared to 44 later events with a correlation of +0.06206 and a slope of +0.04214. Similarly, the EFA stock has a correlation of -0.04610 and the EEM stock has a correlation of -0.05156, which are relatively small and not significant estimates. Thirdly, the SPY stock's range of later 19 trades has a mean of -0.186537%, with a 95% confidence interval of [-0.96363, 0.57048] from 10,000 circular 3-calendar-month blocks, including inactive months, with a seed of 906. The T+1 range of 12 trades has a return of -8.791%, with a mean of -0.75713%, while the prior T+2 has a return of +5.63693%, with a mean of +0.79162%. However, there is no causal attribution to settlement change because the market regime also changed. The EFA stock's range has a return of +0.62029% over 19 trades, and the EEM stock's range has a return of -4.60343% over 21 trades, but neither has a positive lower confidence bound. The conclusion is that there is no reliable positive edge, and the primary range quarter Kelly clips 0. The full hold positive SPY 1.22081% annual 0.3313% is not robust, and the frictionless adjusted buy hold 109.5698% has radically different exposure, not a fair risk-matched alpha test. All three market observations are correlated, and not 57 independent SPY trades. A counterexample is that larger signed-flow shocks may exhibit stronger reversal than the negative-return proxy. The initial event-block bootstrap was replaced with registered calendar-month blocks before the final recording, and the saved stats.json contains the final numbers. The sources of the data are https://runway.adityaperswal.workers.dev/data/stock-bars and the primary paper https://aaltodoc.aalto.fi/server/api/core/bitstreams/0f01d3c2-3b16-44f9-9135-3f03cdd29334/content.
A recent audit checked the accuracy of trading data from December 1-3 for Bitcoin, Ethereum, and Solana. The audit found that daily opening prices were higher than hourly prices by 6.2064%, 7.9068%, and 10.2992% respectively. Daily trading volume was also lower than the sum of hourly volumes by 82.7396%, 89.8276%, and 80.2971%. Although daily closing prices matched hourly closing prices, this does not confirm the accuracy of opening prices, trading ranges, and volumes. A comparison with Coinbase data showed that the opening price of Bitcoin on December 2 was approximately $86,282.35, which is close to the Alpaca hourly price of $86,270.82, but not the daily price of $91,625.13. The audit also checked Litecoin data for 979 calendar days and found four missing hours on three days. The maximum discrepancy in opening prices was 6.55235%, and the maximum difference in daily and aggregated closing prices was 0.240743%. The audit concluded that daily figures cannot be used to support exact stop-size and volume claims, and only hourly data should be used for sensitivity analysis. The audit requires 15-minute executable quotes and prospective evidence before adoption. The sources used for the audit include https://runway.adityaperswal.workers.dev/data/crypto-bars, https://api.exchange.coinbase.com/products/BTC-USD/candles, and https://docs.alpaca.markets/us/docs/historical-crypto-data-1.
The analysis of the GEV payoff is uncertain due to historical data. The study is based on certain assumptions: there are 5 closed holdout trades and no closed training trades. Using a method called NumPy with a seed of 906, the analysis resampled 10,000 times and found the average payoff to be 23.11798%. The range of possible payoffs is between -6.49303% and 58.89909%, which is likely an underestimate. Looking at daily data from 2025 for the SPY stock with 420 aligned returns, the analysis found a beta of 1.78518 and an annual arithmetic intercept of 45.17725%, with a correlation of 0.554865. This data is descriptive, not predictive. The average win of 66.1459% is due to long-term holds and cannot be used to set short-term targets. Two recent fund wins were due to forced liquidations and do not provide a reliable sample. The analysis is cautious and would require a more robust sample with actual risk control to change its eligibility. The data is archived from 2958 to 2997 and is sourced from https://runway.adityaperswal.workers.dev/data/stock-bars. The conclusion is that there is no authorized size that is valid under edgegate and no forecast that GEV will fall.
A study was done to predict the volatility of 14 underlying assets. The study was based on three main ideas: (1) the volatility of an asset in the future affects the value of its long options, (2) a forecast of historical volatility can inform value, but there is a difference between physical and risk-neutral distributions, and (3) it is necessary to consider executable quotes and model uncertainty. The method used daily price data from January 2023 to September 4, 2026, from the IEX exchange and IBIT data from January 2024 onward. The study used a formula to calculate daily returns and variance, and then used a linear regression model to predict future variance. The model was trained on data before December 15, 2025, and then tested on data from January 2026 onward. The results showed that the model performed better than a baseline model for some assets, such as QQQ and SPY, but not for others, such as TLT. The forecast estimates for volatility were 16.3489% for QQQ, 13.3867% for SPY, 21.2358% for IWM, and 54.5551% for PATH. However, the study noted several limitations, including a small sample size and the difference between physical and risk-neutral distributions. The study concluded that the model should be used as a research diagnostic and that no trading edge was established, so the size of the trade should be zero.
An audit was conducted on the consistency of bar data. The data was repaired up to September 4, 2023, at 23 UTC, and included 24,428 rows for BTC and ETH from November 22, 2023, and 17,737 rows for SOL from August 26, 2024, at 16 UTC. There were missing hours on April 4, 7 hours were missing, and the expected holdout was 7824 but only 7823 were found for BTC and ETH, and 7823 for SOL. The largest absolute difference between daily and hourly open prices was found on December 2, 2025, with 6.2064% for BTC, 7.9068% for ETH, and 10.2992% for SOL. The daily and hourly open prices were 91,625.132 and 86,270.82 for BTC, 3,020.1375 and 2,798.84 for ETH, and 139.716 and 126.67 for SOL, respectively. The causes of these differences are unresolved and should not be assumed to be due to management frequency. A comparison of actual holdout entry points found a maximum open difference of 0.467143% for BTC, with one entry differing by more than 10 basis points. From March 10 to March 13, 2026, a timeout win of 2.335254% was found for BTC, but this outcome has an unobserved barrier path and cannot be claimed as an exact stop test. Both daily and hourly implementations resulted in losses, and no current entry preserves rejection. The reported return magnitudes are diagnostics with data uncertainty and not validated performance. An independent cash ledger reconciliation proves accounting given inputs, but not input truth. The primary data can be found at https://runway.adityaperswal.workers.dev/data/crypto-bars. The fields, ranges, missing lists, and comparisons are archived in audit.py and audit2.py. Vendor aggregation reconciliation or independent quote and trade history are needed before future positive adoption. Remaining missing global OI, borrow, and liquidation participant data limit causal claims. No optimistic repair or holdout retuning was performed.
A news scan found the headline about sector reversal uncertainty and passive exposure. The analysis is based on three premises. Premise 1 involves 549 later trades, 184 rotation decisions, and 44 entry months, with a note that treating 549 as independent exaggerates information. A NumPy analysis with 10,000 circular 3-month block resamples, using a seed of 2026090504, retains concurrent trade clusters, with a mean 95% interval of -0.1050% to 0.3364% and a range of -0.2139% to 0.0512%. These are conditional historical intervals, not posterior probabilities. Premise 2 compares equal nine ETF buy-hold later, with a return of 72.671% and a drawdown of 16.389%, to a fixed return of 19.897% and a drawdown of 18.425%. The daily OLS beta is 1.0168, and the annual arithmetic intercept is -9.903%. The range beta is 0.4178, and the intercept is -10.540%, which is descriptive but does not claim causal alpha or significance. The training basket return is 85.453% versus a fixed return of -21.436%, with a range of -16.849%. Premise 3 notes that even the range zero-cost mean CI is -0.0178% to 0.2502%, and the 2-bp one-way range is -0.0561% to 0.2102%, which spans zero. The fixed low-cost mean is positive, but execution and benchmark remain constraints. The conclusion is that there is no robust deployable edge for the range, and the fixed positive headline is not selection alpha. A strong counterargument is that smaller actual costs or turnover netting could improve fixed returns, and repeated forced exit and rebuy overstates optimized turnover. Netting is untested, and not all reversal fails. The current surviving ETF selection, shared research dates, stationarity, and corporate action assumptions limit soundness. Three own forced liquidations provide 0 natural exits, which is not a conditional edge. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, with a query of trades fundquant. Full statistics, symbol, year rows, and seed are in the code.
A news scan found the text describing the historical payoff uncertainty of GRAL, which cannot calibrate the FDA panel. The text includes various statistics, such as a 30.7692% win rate and a -3.03624% mean return. It also mentions a 95% confidence interval of -7.58618% to 1.62648%, which allows for a positive population mean but does not establish it. The text notes that selected factors, regime clustering, and generic price events increase real uncertainty. It also references a daily regression analysis of GRAL and SPY returns from 2026, with a beta of 2.34322 and a correlation of 0.309332. Additionally, the text mentions that the return may reflect stock-specific binaries unrelated to the September panel, and that a catalyst with 0 wins out of 3 closed trades does not estimate panel win chance. The source of the text is https://runway.adityaperswal.workers.dev/data/stock-bars, and it references various files and queries, including tests.json, stats.json, and directquery_tradesfundcatalyst.
A news scan found the headline 'Galleri denominator audit: high specificity and substantial conditional false-positive burden coexist'. The audit results from October 2025 for PATHFINDER 2 initial matrix show the following numbers: true positives were 133, false positives were 83, false negatives were 196, and true negatives were 22,749, with a total of 23,161. This gives a prevalence of 1.4205%, a positive predictive value (which is the probability that a positive test result is correct) of 61.5741%, episode sensitivity (which is the ability of the test to detect cancer) of 40.4255%, and specificity (which is the ability of the test to correctly identify those without cancer) of 99.6365%. The Wilson interval for the positive predictive value at 95% is between 54.9386% and 67.8050%. By May 2026, the full results showed 173 out of 287, or 60.2787%, with a Wilson interval of 54.5176% to 65.7683%. The cohorts overlap, so they should not be treated as independent for difference testing. There were 25,578 initially enrolled versus 23,161 analyzable, which is a 9.4495% difference due to eligibility and follow-up gaps, not automatically a loss to follow-up. The source for this information is https://investors.grail.com/static-files/599b1d07-9463-4a3e-b49f-fa074ab63083 and https://grail.com/press-releases/grail-presents-pathfinder-2-results-of-more-than-35000-participants-showing-the-galleri-test-substantially-increased-cancer-detection-with-robust-performance-and-favorable-safety-at-2026-a/. If the initial sensitivity and specificity are held constant, hypothetical prevalences of 0.5%, 1%, and 2% give positive predictive values of 35.85%, 52.90%, and 69.41%, respectively. This is a Bayes illustration, not a future clinical forecast, as patient mix, test version, and follow-up may change both inputs. A specificity of 99.6% is not the probability that a positive person is cancer-free. The NHS 197146 pooled round observations show true positives of 937, false positives of 864, false negatives of 2114, and true negatives of 193,231, which are not unique people but rather repeated observations. The sponsor's positive predictive value of 52% and episode sensitivity of 30.7% address different conditionals. The first-round false positives of 303 with 54 later diagnoses (44 correct CSO) do not retroactively change the 12-month result. All-screened PF2 invasive results show 213 out of 35,335, or 0.6028%, which differs from the slide 17 conditional of 91% cancer and 50% no-cancer. Denominators are not fully reconciled, and there is no false precise count. NHS AEs recorded collection to referral, PF2 post-lock serious work-up event shows that the ascertainment window matters. The source for this information is https://investors.grail.com/static-files/995005e4-2c8e-4a02-bef0-400403856cc7. The primary p-value of 0.6234 slides versus 0.6324 release is unresolved but both are non-significant, with an incidence rate ratio of 1.03, which is the same. The current registry fixed sequence is confirmed, but the post-result version is not the original SAP. The conclusion is that uncertain utility and conditional burden cannot be compressed into a single approval score. Validity is conditional arithmetic, and soundness is bounded by cohort selection, rounded disclosures, and version differences. Facts are counts, and hypothetical prevalences are explicit assumptions.
A news scan found the headline 'Protein payoff uncertainty and descriptive market dependence'. Premises are based on net natural payoffs from a registered test that has small samples and selected current names. The two live supply wins were operator liquidations and contain no loss estimate or reference quotes. A test using NumPy seed 90504 with 10000 draws and circular 3 trade blocks was conducted. The results showed a 95% mean confidence interval for TSN of -1.4654% to 0.2683%, for PPC of -1.5547% to 0.7230%, and for JBS of -0.6432% to 1.7526%. All of these intervals include zero, which means there is no positive lower bound for risk. Block resampling retains local dependence but cannot remove selection or regimes, or unknown tails. A descriptive analysis of 2025 and later stock daily OLS versus SPY used 420 TSN and PPC days, and 308 JBS days. The betas were 0.16053, -0.04664, and 0.28185. The annual arithmetic intercepts were -2.817%, -7.212%, and -0.659%, but these are not causal alpha or policy effects. The prior is cautious, and a replicated prospective fixed rule with actual venue exits and positive net confidence would change it, but repeating the headline won't. A valid deduction is that zero authorized risk is allowed under fund criteria, and soundness is conditional on historical data. The source of this information is https://runway.adityaperswal.workers.dev/data/stock-bars, which includes observations 1723-1737 and saved raw data and code.
A news scan found the report of an automated trading fund manager. The report analyzed data from October 14, 2025, to September 1, 2026. It looked at the next three days of open-to-open overlapping observations, which is a way to compare data, for 323 assets per type of asset. The report found that extreme negative funding, which is the money paid to hold a position, from the previous day resulted in a gross mean return of -0.6221% for Bitcoin, -0.0827% for Ethereum, and -3.5426% for Solana, compared to other days which were -0.2193%, -0.3447%, and +0.0184%. The report also used a statistical method called circular block bootstrap to analyze the data, which gives a range of possible outcomes. The results showed that the returns overlapped, meaning they were not independent, and the event counts were not independent trades. The report used a statistical model called OLS to analyze the relationship between funding and returns, and found that for Solana, more negative funding was associated with weaker subsequent returns, but this does not mean that one causes the other. The report also looked at the moving average, which is a way to smooth out data, and found that Solana showed a downtrend, or a decrease in value, of -2.12% in 10 out of 50 periods. The report concluded that the evidence favors continued weakness in Solana, but this is not a universal prediction. The report also stated that the current universe, multiple comparisons, and selected regimes limit the ability to make conclusions. The report rejected a fixed rebound rule, which is a rule that predicts a rebound in prices after a decline. The report cited sources from Binance and a crypto data website, and stated that the raw models and data were saved.
The results of a test, called test113, show that the average return on investment was 0.297502% during the training period, with a range of -0.740855% to 1.462600% in 95% of cases. However, during the holdout period, the average return was -0.853688%, with a range of -2.351641% to 0.378149%. This test used a Python program with a random number seed of 20260905 and 20,000 random samples. The test found that the return on investment for Copart was related to the return on investment for SPY, with a beta of 0.613775, which measures how much the return on investment for Copart changes when the return on investment for SPY changes. The intercept was -0.222577%, which means that Copart's return on investment was lower than SPY's by this amount. The test also found that the return on investment for Copart was not stable and was above the maximum allowed by the capital cap. Based on these results, the decision is to reject this implementation, but not the idea that there is an anomaly in the market. The data used for this test came from https://runway.adityaperswal.workers.dev/data/stock-bars and from notices issued by the company during certain periods.
A news scan found a report about a trading strategy. The strategy is not a new optimized strategy, but rather a benchmark falsification. The strategy involves buying 19 country ETFs equally at the start of a period, charging a 5 basis point fee and 5 basis point slippage, and holding them until the end. The daily adjusted closes are marked with no terminal sale. From 2017 to 2022, the strategy resulted in a 31.9001% return, with a 39.1696% drawdown. From 2023 to September 4, the strategy resulted in a 101.8003% return, with a 14.0588% drawdown. A monthly rotation strategy resulted in a 113.2853% return, but with higher risk, including a 19.2653% drawdown and a daily OLS beta of 1.23544. The strategy was compared to the SPY, which had a 3.00% intercept in analysis, indicating that the choice of benchmark materially changes the interpretation of the results. The report concludes that while the strategy may be profitable, it is not proven to be superior to passive exposure, and its validity is descriptive only. The report notes that the global equity exposure itself earned substantial returns, and that turnover and concentration add risk. The current country sample is not a comprehensive index universe and survives to today. The report uses a formula to calculate daily values, which is q_s = (10000/19) / (first_open_s * 1.0005^2), and daily value = sum(q_s * close_s), with OLS regression used to analyze the results. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars and artifacts/benchmark.json.
A country momentum dependence audit was conducted. The audit analyzed 130 completed monthly trades and 133 range trades from September 4, 2023, to 2026. The trades were grouped by entry month, resulting in 44 and 45 clusters. Using NumPy with a seed of 90502, 10,000 independent month-cluster resamples were generated, producing mean-return 95% intervals of 0.21096% to 3.50982% for monthly trades and -0.56978% to 0.72463% for range trades. A stronger serial-dependence sensitivity test was also conducted with a seed of 90503, using stationary blocks that averaged three entry months, and 10,000 samples, resulting in mean-return 95% intervals of 0.35802% to 3.46067% for monthly trades and -0.51449% to 0.60859% for range trades. The positive fractions of 0.9938 and 0.6264 are bootstrap frequencies, not current win probabilities. Daily ordinary least squares analysis against the SPY ticker showed a monthly beta of 0.93818 and an annual arithmetic intercept of 3.00129%, with an R2 of 0.38703, while the range beta was 0.06893 and the annual arithmetic intercept was -0.53141%, with an R2 of 0.01884. The correlation between EWY and EWT was 0.76154, between EWY and EWP was 0.47974, and between EWT and EWP was 0.52041. The monthly annual returns were 14.83% for 2023, -8.26% for 2024, 40.23% for 2025, and 44.37% for 2026 to date, while the range returns were 3.25%, -2.08%, 6.00%, and -4.38% for the same periods. The conclusion is to distinguish between positive monthly historical evidence and failed range implementation, and not to reject all momentum or use monthly payoff to size range exits. The soundness of the audit is bounded by current surviving wrappers, one unusually strong recent chip cycle, historical rather than prospective holdout, and monthly dependence beyond resampling assumptions. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the scripts used are artifacts/stats.py and monthblock.py, with backtests 70-77.
A news scan found the following text: Premise 1 uses 419 complete common daily return rows from September 4, 2025, from all-adjusted IEX. This compares ordinary least squares, or OLS, versus SPY beta for several stocks: AXTI is 2.511, AEHR is 3.447, PSIX is 2.892, VSH is 2.387, POWL is 1.899, and ETN is 1.539. The correlation between POWL and ETN is 0.6593, and between AEHR and VSH is 0.5628. This shows that distinct products do not eliminate shared market exposure. The intercepts are descriptive, not causal alpha, and there is no annualized alpha claim. Premise 2 uses a mean of 10,000 circular three-trade block bootstrap seed 905 on natural exit samples of 40, 39, 29, 38, 34, and 32. Five means have a 95% interval crossing zero. The mean for PSIX is 3.7507%, with a confidence interval of 0.4832% to 7.1587%, but this is based on only four training bars, which is the selected post-listing history. An additional 20,000 draws give a conservative six-test tail sensitivity of -0.6271% to 8.3548%. This cannot correct all 3012-name screening and unknown prior research, and is not a formal population guarantee. Removing the best PSIX trade mean of 2.5700% shows that it is not entirely one outcome, and there is still small sample and regime risk. Premise 3 notes that contemporary survivorship, selected high movers, assumed costs, conditional barrier fills, and missing natural ledger losses undermine transportability. The conclusion is that PSIX merits prospective collection, not size, and others are rejected or watched per backtests 83-88. The source is https://runway.adityaperswal.workers.dev/data/stock-bars, with stats.json, tests.json, and derived.json/trades archived. The prior on PSIX shifts from unknown to interesting diagnostic, but was never adopted, and genuinely independent net execution outcomes would change the verdict.
A data analysis from the source https://runway.adityaperswal.workers.dev/data/crypto-bars found that from January 1 to September 4, 2026, the net natural-exit samples for the cryptocurrencies XRP, LINK, UNI, and PAXG were 6, 14, 13, and 8, respectively. Using 20,000 independent and identically distributed resamples with a seed of 2026090502, the mean trade 95% intervals were calculated as XRP [-1.194, 18.765]%, LINK [-1.726, 2.184]%, UNI [-2.933, 2.634]%, and PAXG [-1.402, 0.419]%. This analysis assumes that the trades are independent, but the current selection of trades and market conditions may limit the interpretation of the results. The frequencies of positive resamples were 94.375% for XRP, 51.585% for LINK, 38.65% for UNI, and 13.8% for PAXG, but these are not posterior probabilities. A daily ordinary least squares analysis on BTC with an intercept found betas of 1.237, 1.176, 1.181, and 0.299 for XRP, LINK, UNI, and PAXG, respectively, with correlations of 0.871, 0.874, 0.677, and 0.446. The sample size was 247 for each cryptocurrency except PAXG, which had a sample size of 204. The similar betas for XRP and LINK weaken the narrative of diversification. The raw binary mean-payoff quarter Kelly notional formula gave capital fractions of 3.889, 0.281, -0.468, and -7.550, which are clearly unstable and constrained before caps. Since the lower mean bounds are negative, the authorized allocation is 0. The premises of a positive XRP return and a win rate of 66.67% depend on only 6 trades, and the strongest bull case is a large payoff, but no precise population edge has been established. There is a need for prospective fixed rule evidence and tested venue exits, and one should not pick the best lookback from the inspected holdout. The UNI terminal open mark was excluded from payoffs rather than counted as a winner.
A news scan found the data from a crypto market analysis. The data, available at https://runway.adityaperswal.workers.dev/data/crypto-bars, covers daily returns from January 1 to September 4, 2026. It shows that the correlation between Bitcoin (BTC) and Ethereum (ETH) is 0.9124, between BTC and Solana (SOL) is 0.8805, and between ETH and SOL is 0.8801. This means that these currencies tend to move together in the market. The beta values, which measure volatility, are 1.254 for ETH and 1.202 for SOL, based on ordinary least squares (OLS) regression, a statistical method. The daily standard deviation, a measure of risk, is 2.452% for BTC, 3.370% for ETH, and 3.347% for SOL. The worst close-to-close return, which is the change in price from one day to the next, was -13.986% for BTC, -14.890% for ETH, and -14.931% for SOL. Two premises are considered: first, that the three major long positions are strongly correlated and represent a risk, rather than independent investments; and second, that the registered net daily closed-trade bootstrap, a statistical method, shows mean 95% intervals of -1.424% to 1.323% for BTC, -2.670% to 0.678% for ETH, and -1.213% to 6.606% for SOL. For SOL, there were 12 trades, with a 50% win rate, a mean win of 6.123%, and a mean loss of 2.270%. The naive quarter-Kelly, a betting strategy, is 0.07866, but the authorized size is 0 because the interval crosses 0, and there is no prospective validation or tested venue exits. The bootstrap-positive fraction of 0.7897 indicates resampling stability, but not the probability that the strategy is profitable. The assumption of independent and identically distributed (IID) data ignores serial dependence, which may lead to understated uncertainty. All hourly net mean intervals were entirely negative, ranging from -0.611% to -0.443% for BTC, -0.609% to -0.346% for ETH, and -0.682% to -0.465% for SOL. Despite this, SOL may have a real positive momentum edge, but there is insufficient confidence to prove a negative return. The goal is to achieve repeatable net gains, with the constraint of independent payoff validation. The methods and results are available in the return-statistics.json, bootstrap.json, and risk_analysis.py files, but no predictive regression is claimed.
A study of tanker stocks found that the relationship between two stocks, FRO and STNG, is strong, with a Pearson correlation of 0.7506 from January 2025 to September 2026. This means that the prices of these two stocks tend to move together. The study also found that the stock return on SPY, a common stock market index, plus an intercept, has slopes of 0.6291 and 0.4025. This suggests that even though the market beta, a measure of volatility, is low, the shared exposure to tanker stocks is not. The study looked at 41 and 38 trades for FRO and STNG, respectively, and found mean returns of 0.53025% and 0.37899%. However, these numbers are not reliable because they are based on a small number of trades and do not account for average losses. The study also used a technique called resampling to estimate the range of possible returns for these stocks. The results showed that the mean returns for FRO and STNG could be between -0.87436% and 1.95854%, and -0.75304% and 1.45383%, respectively. The study concluded that the evidence is not strong enough to say that the mean return is positive, and that the original idea of diversifying between these two stocks is not supported. The study also noted that the results are limited by the small number of trades and the fact that the market conditions may have changed. The study did not claim to have found any statistically significant results. The data for the study was retrieved from https://runway.adityaperswal.workers.dev/data/stock-bars on September 5, and the code and notes are archived.
A descriptive analysis of 18 matched 9-session event returns for CHWY was done using a statistical method called ordinary least squares, or OLS for short, which is a way to find the best fit line for a set of data. The results showed that net CHWY returns can be represented by the equation netCHWY = alpha + beta * raw SPY, where alpha is -4.33%, beta is 3.02, correlation is 0.848, and R2 is 0.720. This means that CHWY returns tend to move with the market, represented by the SPY ticker. The holdout mean CHWY returns minus SPY returns were -0.611%. No filter was used to select trades, so the relationship between CHWY and SPY returns could be due to market conditions or unusual events, rather than a consistent premium. The analysis had some limitations, including only looking at one issuer and announced dates, rather than the full market. Six out-of-sample events had a mean return of -0.153%, with a 95% confidence interval of -4.53% to 3.69%. The results from the fund's own trade ledger showed 0 wins, a mean loss of 16.38%, and a total loss of -31.94, with one rare stop and two operator exits. The conclusion is that there is no numerical evidence of a positive edge, or advantage, in trading CHWY. The validity of the results is limited by the small and selected sample size. A news scan found the source of the data to be https://runway.adityaperswal.workers.dev/data/stock-bars and a query of trades from the catalyst fund on September 5. The distinction between rate and total is important, as a mean return of -0.153% per holdout window produces a compounded return of -1.72% across 6 events, not six times the mean return, due to the effect of volatility drag. Prior cash levels were unchanged, and only a preregistered prospective positive net sample with implementable exits would change eligibility, not the monthly shortfall. A gap loss of -6.44R versus the intended stop shows that normalized stop-risk alone can be misleading.
A news scan found the headline is not available, but a report on global trend holdout uncertainty and beta was released. The report discusses defense without demonstrated alpha. It mentions SIP-derived 2023-2026-09-04 fixed 10-month returns, which came from https://runway.adityaperswal.workers.dev/data/stock-bars. The report states that 18 closed trades had an average return of +2.3818%. A bootstrap analysis, which is a statistical method to estimate uncertainty, was done with 20,000 repetitions and a seed of 20260905. The mean return was between -1.1377% and 6.3696% with 95% confidence. About 10.375% of the draws were nonpositive. The report notes that trades overlap across correlated markets, so the interval is only diagnostic. If the best trade, which had a return of +24.5168%, is removed, the mean return falls to +1.0797%. A joint stationary block bootstrap, which is another statistical method, was done with 20 sessions and 3,000 full-horizon paths. The annual arithmetic strategy mean return was between 1.8463% and 13.6657%. The active versus benchmark return was between -9.7113% and -0.7710%. The report states that this is conditional resampling of one history, not a population confidence guarantee. A daily ordinary least squares analysis, which is a statistical method to model the relationship between variables, was done versus an equal-initial-sleeve buy-hold. The beta, which measures volatility, was 0.6631, and the annualized intercept was -0.7287%. The R2, which measures the goodness of fit, was 0.7887. The report notes that this is descriptive, not causal, and does not imply alpha, which refers to excess return. The daily correlation between SPY and EFA was 0.7598, and between SPY and IEF was 0.0528 in the test period, versus 0.1931 in 2022. The report gives a counterexample, stating that observed lower drawdown can follow reduced exposure alone, and a different crisis may break diversification. The validity of the report is limited by the small number of trades, and the soundness is limited by the selected asset-class wrapper set and missing 2008 data. The report preserves the 10-month rule and does not select 12 months after looking. The policy allocation is 0.
A study of 7 events with 20-day outcomes found that the relationship between the opening gap and the 20-day return of the SPY stock, adjusted for earnings per share surprise, had a gap coefficient of -1.45 and a surprise coefficient of +0.0083, with an R-squared value of 49.7%. The correlation between the gap and return was -0.703, and the correlation between the log of the earnings surprise and return was -0.155. However, these estimates are only descriptive and not reliable due to a small sample size of 7, inconsistent data, and conflicting definitions of earnings per share. For example, DELL's stock continued to rise by 28.0% over 20 days after a 13.1% gap in February, while HPE's stock reversed and fell by 28.1% after a 33.7% gap. The study concludes that neither a larger gap nor a larger earnings surprise supports a stronger expected drift, and if anything, the gap may indicate exhaustion, with no edge in sizing.