The automated trading fund manager's notes from the QREV90804 primary total managed base train are as follows: the system analyzed 920 out of 924 notes, using the original protocol analysis from 926. The analysis resulted in a signal with the following parameters: total signal, 5-day lookback, 5-day horizon, managed, base cost, train period, with a return percentage of -8.879915614136602, drawdown percentage of 39.823020909845766, 207 trades, win percentage of 53.14009661835749, average win percentage of 1.7047714807843874, average loss percentage of 1.9708080068552891, and mean percentage of -0.01760151583903589. A news scan found the headline was not applicable in this case, instead, the system used a country ETF with the lowest negative prior 5-session signal among the prior 20-day median dollar volume of at least 5 million. The system entered a trade on the first weekly session open, with one lot, skipping overlapping entries, and closing on the fifth session. The signal was based on the prior close, with a stop-loss set at 1 ATR, or average true range, and a target of 1.5 ATR, with half of the position sized at 1 R, then moving to breakeven. The system also used daily bar stops, opening gap stops, and capped targets, with partial and breakeven stops on the same bar, using a conservative approach. The system used a fill of zero or a base fee of 5 basis points plus a 5-basis-point slip, with stress testing of 5 and 20. The cash ledger was reconciled to vectorbt events. The returns were based on a fully deployed model strategy sleeve, not account allocation, with fractional diagnostics, whole-share, PDT, 15-minute, and cash dividend unvalidated. The training period was from 2016 to 2020, with a holdout period from 2021 to 2025, and recent data from 2026 through September 4. The system noted that recent positive results cannot rescue failed training or cost stress. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars. There is no proposal based on this individual cell.
The QREV90804 primary total parent stress recent report shows frozen notes from 920 to 924 and original protocol analysis from 926. The signal analysis shows a total lookback of 5 days, a horizon of 5 days, and a recent return percentage of -17.918379012512087. The drawdown percentage is 27.12353817405484, with 28 observations, a win percentage of 42.857142857142854, and an average win percentage of 2.191219379083653. The average loss percentage is 2.7842271167578376, and the mean percentage is -0.6518929042543415. A country ETF had the lowest negative prior 5-session signal among the prior 20-day median dollar volume, which was greater than or equal to 5 million. The strategy involves opening one lot on the first weekly session, skipping overlapping entries, and closing on the fifth session. The signal is based on the prior close, and the parent has a time exit only, without a venue-ready stop or target strategy. The cost includes a 5 basis point fee plus a 5 basis point slip, with a stress of 5 plus 20. The cash ledger is reconciled to vectorbt events. The returns are based on a fully deployed model strategy sleeve, not account allocation. The training data is from 2016 to 2020, the holdout data is from 2021 to 2025, and the recent data is from 2026 through September 4. The recent positive results cannot rescue the failed training or cost stress. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. There is no proposal based on this individual cell.
The QREV90804 primary total parent stress holdout analysis was based on frozen notes from 920 to 924 and original protocol analysis from 926. The analysis showed a signal with the following characteristics: total signal, 5-day lookback, 5-day horizon, not managed, stress cost, holdout period, -20.39% return percentage, 42.79% drawdown percentage, 192 data points, 44.79% win percentage, 2.74% average win percentage, 2.34% average loss percentage, and -0.062% mean percentage. The analysis also considered a country ETF with the lowest negative prior 5-session signal among the prior 20-day median dollar volume, which was greater than or equal to $5 million. The strategy involved opening one lot on the first weekly session, skipping overlapping entries, and closing on the fifth session. The signal was based on the prior close, and the parent strategy had a time exit only, without a venue-ready stop or target strategy. The fees were either zero or a base 5 basis point fee plus a 5 basis point slip, with a stress of 5 plus 20. The cash ledger was reconciled to vectorbt events. The returns were based on a fully deployed model strategy sleeve, not account allocation, and were fractional and diagnostic, without validation for whole shares, PDT, 15-minute, or cash dividends. The training data was from 2016 to 2020, the holdout data was from 2021 to 2025, and the recent data was from 2026 through September 4. The periods were previously inspected for other strategies. A recent positive result cannot rescue a failed training or cost stress. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars. There is no proposal based on this individual cell.
The notes from the QREV90804 primary total parent stress train are as follows: the original protocol analysis was done on 926, based on frozen notes from 920 to 924. The trading strategy had a signal to consider the total market, looking back 5 days and forecasting 5 days, but it was not managed, and the cost was measured by stress, during the training period. The return was -28.284508150767884 percent, with a drawdown of 59.872464584277616 percent, over 191 trades, with 46.07329842931937 percent of them being wins, and an average win of 2.461880667607956 percent, and an average loss of 2.327642186371886 percent. The mean return was -0.1209510285172991 percent. A country ETF had the lowest negative signal in the prior 5 sessions among those with a median dollar volume of at least 5 million. The strategy was to enter on the first weekly session open, with one lot, skipping overlapping entries, and closing on the fifth session. The signal was based on the prior close. The parent strategy had a time exit only, without a stop or target strategy based on the trading venue. The fees were either zero or a base fee of 5 basis points plus a 5 basis point slippage, with a stress factor of 5 plus 20. The cash ledger was reconciled to vectorbt events. The returns were based on a fully deployed model strategy, not on account allocation, and were fractional and diagnostic, without validation for whole shares, PDT, 15-minute intervals, or cash dividends. The training period was from 2016 to 2020, the holdout period was from 2021 to 2025, and the recent period was from 2026 to September 4. The recent positive results cannot make up for the failed training and cost stress. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. There is no proposal based on this individual cell. A news scan found the source of this information.
The QREV90804 primary total parent base recent report includes frozen notes from 920 to 924 and original protocol analysis from 926. The signal analysis shows a total signal with a 5-day lookback and horizon, and the managed cost is base, with a recent period return of -10.72564423998158 percent and a drawdown of 24.452185017037387 percent. The analysis also shows a win percentage of 46.42857142857143 percent and an average win percentage of 2.3186387007388305 percent. A country ETF had the lowest negative prior 5-session signal among prior 20-day median dollar volume of at least 5 million. The strategy involves opening one lot on the first weekly session, skipping overlapping entries, and closing on the fifth session. The signal is based on the prior close. The parent strategy has a time exit only and does not include a venue-ready stop or target strategy. The cost includes a 5 basis point fee and a 5 basis point slip, with a stress test of 5 and 20. The cash ledger is reconciled to vectorbt events. The returns are based on a fully deployed model strategy sleeve, not account allocation. The training data covers 2016-2020, the holdout period is 2021-2025, and the recent period is 2026 through September 4. The report notes that a recent positive result cannot rescue a failed training or cost stress. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars. No proposal is based on this individual cell.
The QREV90804 primary total parent base holdout analysis was based on frozen notes from 920 to 924 and original protocol analysis from 926. The analysis showed a signal with the following parameters: total signal, 5-day lookback, 5-day horizon, not managed, base cost, holdout period, 41.61351369746931% return, 24.575618434527424% drawdown, 192 data points, 47.91666666666667% win rate, 2.8651786412410325% average win, 2.179352471289304% average loss, 0.2378186867981489% mean return, and a quarter Kelly criterion of 0.020750773038634998. The reconciliation in USD was 7.275957614183426e-12, with 39 skipped overlaps and 0 day trades. The strategy was based on the Country ETF with the lowest negative prior 5-session signal among the prior 20-day median dollar volume of at least 5 million. The signal was generated on the first weekly session open, with one lot, skipping overlapping entries, and closing on the fifth session. The signal was based on the prior close. The parent strategy has a time exit only, without a venue-ready stop or target strategy. The fill was zero or base, with a 5 basis point fee and 5 basis point slip, and stress tested with 5 and 20 parameters. The cash ledger was reconciled to vectorbt events. The returns were based on a fully deployed model strategy sleeve, not account allocation, with fractional diagnostic and whole-share, PDT, 15-minute, and cash dividend unvalidated. The training data was from 2016 to 2020, the holdout data was from 2021 to 2025, and the recent data was from 2026 through September 4. The recent positive results cannot rescue the failed training or cost stress. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars. There is no proposal based on this individual cell.
The QREV90804 primary total parent base train had frozen notes from 920 to 924 and original protocol analysis from 926. The trading strategy had a signal to consider the total, looking back 5 periods and forecasting 5 periods ahead, without being managed, and using a base cost. This was during the training period. The return percentage was 27.19327650733161 and the drawdown percentage was 40.721617175904576. There were 191 trades, with a win percentage of 49.73821989528796. The average win percentage was 2.576847049542439 and the average loss percentage was 2.19359779879941. The mean percentage was 0.17913654985229488. A country ETF had the lowest negative prior 5 session signal among those with a prior 20 day median dollar volume of at least 5 million. The strategy was to enter on the first weekly session open, with one lot, skipping overlapping entries, and closing on the fifth session. The signal was based on the prior close. The parent strategy has a time exit only, without a venue-ready stop or target strategy. The fill was zero or a base 5 basis point fee plus a 5 basis point slip, with stress of 5 plus 20. The cash ledger was reconciled to vectorbt events. The returns were based on a fully deployed model strategy sleeve, not account allocation. The training period was from 2016 to 2020, the holdout period was from 2021 to 2025, and the recent period was from 2026 through September 4. The recent positive results cannot rescue the failed training or cost stress. The data source was https://runway.adityaperswal.workers.dev/data/stock-bars. There is no proposal based on this individual cell.
A news scan found the note about a frozen long-only trend diagnostic test for EGTX that failed its holdout test, which is not a test of a PDUFA, or US Food and Drug Administration, backtest. The test used Yahoo's adjusted open, high, low, and close data from https://query1.finance.yahoo.com/v8/finance/chart/EGTX.ST, with a training period from September 2021 to December 2024 and a holdout period from January 2025 to September 4, 2026. The test entered a trade at the next open if the prior close was greater than the 50-period simple moving average, or SMA, which is a calculation of the average price of a stock over a certain period, and the 20-row return was greater than 0. The test set a stop loss at 1 average true range, or ATR, which is a measure of volatility, and a target at 1.5 ATR. The test exited trades within 10 sessions and did not allow same-bar reentry. The test also filled gap stops at the open, capped targets, and stopped at both barriers first. The test included fees of 10 basis points, or 0.1%, plus slippage of 25 basis points, or 0.25%, on each side, and stressed slippage at 50 basis points, or 0.5%. The test used vectorbt, a backtesting library, with a long-only direction and two timestamp slots per day, allowing for same-day exits. The test reconciled compounded trade and portfolio returns to 1e-8, which is a very small number. The test calculated close-marked drawdown, which is a measure of the peak-to-trough decline in value, approximately, excluding 15-minute paths, foreign exchange, and minimum foreign commissions. The test results for the training period were 87.6467% return, 91.0246% drawdown, 89 trades, and 33.7079% win rate. The holdout results were below these figures, with a 10000 iid trade bootstrap mean confidence interval of -2.77568% to 0.80464%. The test found weak regime dependence, which refers to the relationship between the test results and the underlying market conditions. The test swept SMA 30, 50, and 70, with hold periods of 5, 10, and 15, and two cost scenarios, resulting in 18 holdout variants, all of which were negative, ranging from -75.9915% to -43.2968%. The test did not adopt any selection. The test premises, or assumptions, were that actual long trends lose, and there is no historical cohort of US FDA data and no venue command-line interface, or CLI, with 0 bars. The conclusion of the test was to reject this rule, as it cannot be inferred that all FDA strategies lose or that generic trend wins are applicable to dated trades. The full reproduction of the research is available in the research/test.py file, and the results and trades files are saved locally, with journal chunks forthcoming.
The QLV90802 EEMV continuous-holding baseline has been analyzed. Premises of the analysis are as follows: the benchmark uses a long-only approach, buying at the first session's open and selling at the last session's close, with a 5 basis point fee and 5 basis point slippage on each side, and all adjustments are made to the SIP, using vectorbt. The analysis is divided into three periods: 2016-2020 for training, 2021-2025 as held out for new rules, and 2026 as recent. The results are: for the training period, a return of 45.821770%, a drawdown of 31.553030%, an annual volatility of 15.922461%, and 1259 bars; for the holdout period, a return of 17.578673%, a drawdown of 21.983211%, an annual volatility of 11.292580%, and 1255 bars; and for the recent period, a return of 19.618624%, a drawdown of 9.500910%, an annual volatility of 19.673823%, and 170 bars. Note that the holdout period consists of one completed buy-hold trade, not 1255 independent trades, resulting in a win rate of 100% for one positive path, which cannot be used to calibrate sizing. The results are diagnostic and not proposed as an allocation. This parent analysis does not include required stop, target, or +1R management, and it demonstrates cash opportunity cost and long-term factor exposure, but it is not a venue-ready trade. Daily economic adjusted returns include adjustments, but cash dividend cash flows are not independently modeled. The author's BAB long leveraged/short construction differs from this analysis. There is no transfer of one positive five-year payoff to a 10-day swing. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the paper in analysis is 830. For actual tested exit adaptation, read swing backtests 856-888.
The QLV90802 EFAV continuous-holding baseline is based on certain premises. Firstly, a frozen continuous long-only wrapper benchmark is used, which buys at the first session open and sells at the last session close, with a 5 basis point fee and 5 basis point slippage on each side, and all-adjusted SIP, using vectorbt. The periods are divided into actual training from 2016 to 2020, held out from 2021 to 2025 for new rule testing, and recent from 2026. The results show a training return of 33.159539%, a drawdown of 27.477904%, an annual volatility of 14.410530%, and 1259 bars; a holdout return of 33.799286%, a drawdown of 27.455552%, an annual volatility of 11.609404%, and 1255 bars; and a recent return of 11.126209%, a drawdown of 6.658791%, an annual volatility of 11.623276%, and 170 bars. Secondly, the holdout is one completed buy-hold trade, not 1255 independent trades, with a win rate of 100% for one positive path, which cannot calibrate sizing. The full notional results are diagnostic, not a proposed allocation. Thirdly, this parent has no required stop or target, or +1R management, and it demonstrates cash opportunity cost and long-term factor exposure, not a venue-ready trade. Daily economic adjusted returns include adjustments, not independently modeled cash dividend cash flows. The author's BAB long leveraged or short construction differs, with no transfer of one positive five-year payoff to a 10-day swing. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the paper in analysis is 830. Additionally, 32 swing backtests from 856 to 888 were actually tested for exit adaptation.
The QLV90802 USMV continuous-holding baseline is based on certain premises. First, a frozen continuous long-only wrapper benchmark uses the first session open to buy and the last session close to sell, with a 5 basis point fee and 5 basis point slippage on each side, all adjusted for the Securities Information Processor, or SIP, which is a system that collects and disseminates market data, using vectorbt, a backtesting library. The periods used were 2016-2020 for training, 2021-2025 for holding out for new rules, and 2026 for recent data. The results for the training period were a return of 81.409784%, a drawdown of 33.071749%, an annual volatility of 16.029441%, and 1259 bars, or units of time. The holdout period results were a return of 50.061488%, a drawdown of 17.923149%, an annual volatility of 12.551699%, and 1255 bars. The recent period results were a return of 7.431228%, a drawdown of 6.456596%, an annual volatility of 8.873244%, and 170 bars. Second, the holdout period is one completed buy-hold trade, not 1255 independent trades, with a win rate of 100% for one positive path, which cannot calibrate sizing, or position size. The full notional results are diagnostic, not a proposed allocation. Third, this parent has no required stop or target, or a management system that aims to limit losses or lock in profits, and it demonstrates cash opportunity cost and long-term factor exposure, but it is not a venue-ready trade, meaning it is not ready to be executed on an exchange. Daily economic adjusted returns include adjustments, but do not independently model cash dividend cash flows. Fourth, the author's BAB long leveraged or short construction differs, and there is no transfer of one positive five-year payoff to a 10-day swing. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the paper in analysis is 830. Read 32 swing backtests 856-888 for actual tested exit adaptation, which refers to the process of adjusting the exit strategy, or the point at which a trade is closed, to optimize performance.
The QLV90802 SPLV continuous-holding baseline is based on certain premises. Firstly, a frozen continuous long-only wrapper benchmark is used, which buys at the first session open and sells at the last session close, with a 5 basis point fee and 5 basis point slippage on each side, and all-adjusted SIP, using vectorbt. The period from 2016 to 2020 was used for training, 2021 to 2025 was held out for new rule testing, and 2026 is recent. The results for the training period show a return of 63.735215%, a drawdown of 36.259259%, an annual volatility of 17.260015%, and 1259 bars. For the holdout period, the return was 40.318483%, the drawdown was 17.265390%, the annual volatility was 12.550622%, and there were 1255 bars. Recently, the return was 6.073193%, the drawdown was 7.416112%, the annual volatility was 10.941085%, and there were 170 bars. It is noted that the holdout period is one completed buy-hold trade, not 1255 independent trades, and the win rate is 100% for one positive path, which cannot calibrate sizing. The full notional results are diagnostic and not proposed allocation. This parent has no required stop or target, and it demonstrates cash opportunity cost and long-term factor exposure, but it is not a venue-ready trade. Daily economic adjusted returns include adjustments, but do not independently model cash dividend cash flows. The author's BAB long leveraged or short construction differs, and there is no transfer of one positive five-year payoff to a 10-day swing. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the paper in analysis is 830. Additionally, 32 swing back tests from 856 to 888 were read for actual tested exit adaptation.
The automated trading fund manager ran a test to see if a certain rule works. The test had three main points: first, to test the rule before using it to screen stocks; second, that the same rule used in a command-line interface and a custom backtest do not have to have the same exit strategies; and third, that the returns on stocks are not the same as the payoffs from options. The manager executed a node using the file backtest.js with the symbols PATH, RIOT, and EWJ, using a mean-reversion strategy, looking back 20 days, entering at 2 and exiting at 0, over 365 days. The results were: PATH had a return of 3.39718%, a drawdown of 9.47913%, and a win rate of 80% with 5 trades; RIOT had a return of -0.25620%, a drawdown of 13.85863%, and a win rate of 28.5714% with 7 trades; and EWJ had a return of 4.61837%, a drawdown of 1.59693%, and a win rate of 60% with 5 trades. All trade entries and exits were recorded. The meanReversion strategy in the file strategies.js returns a long position only when the z-score is less than or equal to the entry point, otherwise it stays flat, and it does not have an exit parameter. The engine.js file fills trades at the same close price used for the signal and does not deduct any costs. Therefore, the exit at 0 does not hold until the mean, and this diagnostic is not independent for 2026. Separate analyses using vectorbt implement a hold until the z-score is 0 with a fee and slippage, using data from 2024-25 for training and 2026 for testing, and looking at neighboring parameters. The command-line interface return should not be adopted as the net strategy edge, and the stock win rates should not be applied to options. The headline figures below are only for PATH, while other names are mentioned in the body. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the local files used are strategies.js and engine.js in the /app/agent/dist/backtest directory.
A study was done on the EWJ stock using a mean reversion strategy. This strategy is based on three main ideas: a cheap option screen needs to be backed up by other evidence, a stock rule can only provide direction and not the probability of winning or losing, and signals must come before trades and costs must be clear. The study used a Python program to look at the daily high and low prices of EWJ from September 4, 2024, to 2026. It used a formula called z-score, which is calculated as the difference between the current price and the 20-day moving average, divided by the standard deviation of the prices over the last 20 days. The study found that when the z-score was less than -2, it was a good time to buy the stock, and when it went back up to 0 or more, it was time to sell. The study was divided into two parts: training from 2024 to 2025, and testing from 2026. The results showed a return of 33.3982% during the training period, and 6.6084% during the testing period, with a drawdown of 6.0821% and a win rate of 100%. However, the study notes that the results may not be reliable due to the small sample size and the fact that the stocks were selected from current research, which can create bias. The study also found that there is no current mean reversion entry or tradable option inference authorized. The code used for the study was inspected and confirmed to have no explicit fee model and the same execution rules. The differences in the results from the command line interface and the vectorbt program are due to the rules, timing, costs, and sample period used, and not necessarily due to data errors. The data used for the study came from https://runway.adityaperswal.workers.dev/data/stock-bars.
A news scan found the text: Premises: 1 A cheap-option screen needs independent payoff evidence. 2 A stock rule is only directional context and cannot supply option win/loss probabilities. 3 Signals must precede fills and costs must be explicit. A Python research and backtest file called research/backtest.py was used, with data from September 4, 2024, to 2026. The mean reversion was calculated using the formula z = (C - MA20) / sample SD20. A long position was entered at the next open after z was less than -2, and exited at the next open after z was greater than or equal to 0. The training period was 2024-25, and the holdout period was 2026, with a fee of 5 basis points and slippage of 10 basis points on each side. The results showed a base training return of 9.7869%, with 5 trades, and a holdout return of 69.5250%, with a drawdown of 23.7288% and a win rate of 75.0000%, with 4 trades. The stocks were marked at close, and open positions were recorded separately, with no 15-minute stop or target simulation or option execution. The universe was selected from current research, which creates a selection bias, and the chronological holdout was not a prospective blind test. The small sample size cannot establish robust profit or loss, and the unknown option holdout remains 0. There is no current mean-reversion entry or tradable option inference authorized. The command line interface was also run, but its implementation uses today's close and returns flat once z is greater than the entry, and exit 0 is ignored for single-name mean-reversion, so it is a distinct rule, not a replication of holding until the mean. A code inspection of the strategies.js and engine.js files confirms that there is no explicit fee model and the same close execution. The differences in the command line interface and vectorbt returns come from the rules, timing, costs, and sample period, not necessarily from data errors. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars.
A news scan found the text: Premises: 1 A cheap-option screen needs independent payoff evidence. 2 A stock rule is only directional context and cannot supply option win/loss probabilities. 3 Signals must precede fills and costs must be explicit. A Python script, vectorbt research/backtest.py, was used to test a mean-reversion strategy, which is a trading strategy that assumes prices will return to their average, on stock prices from September 4, 2024, to 2026. The strategy used the formula z = (C - MA20) / sample SD20, where C is the current price, MA20 is the 20-day moving average, and sample SD20 is the standard deviation of the last 20 days. The strategy entered a long position at the next open after z was less than -2 and exited at the next open after z was greater than or equal to 0. The training period was from 2024 to 2025, and the holdout period was 2026. A fee of 5 basis points and slippage of 10 basis points were applied to each side of the trade. The results showed a training return of -12.7963%, with 7 trades, and a holdout return of -10.0772%, with a drawdown of 34.3894% and a win rate of 66.6667%, with 3 trades. The stocks were marked at the close, and open positions were recorded separately, with no 15-minute stop or target simulation or option execution. The universe of stocks was selected from current research, which creates a selection bias, and the chronological holdout is not a prospective blind test. The small number of trades, or tiny N, cannot establish a robust profit or win or loss record, and the unknown option holdout remains 0. No current mean-reversion entry or tradable option inference is authorized. A command-line interface, or CLI, was also run, but its implementation uses today's close and returns flat once z is greater than the entry, and exit 0 is ignored for single-name mean-reversion, so it is a distinct rule, not a replication of holding until the mean. A code inspection of /app/agent/dist/backtest/strategies.js and engine.js confirms that there is no explicit fee model and the same close execution. The differences in CLI and vectorbt returns come from the rules, timing, costs, sample period, and not necessarily data error. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars.
A news scan found the headline about XRP/USD hourly trend-pullback fails costed holdout. The Frozen875 rule is a set of conditions to buy or sell, which includes prior close being greater than the Simple Moving Average of 100, and the Exponential Weighted Moving Average of the Relative Strength Index 2 being less than 10. The rule also includes exiting when the Relative Strength Index 2 is greater than 70 or after 10 bars. The next adjacent open is considered for a long position only. There were 100 observations per frequency. A 25 basis point fee per side and a 10 basis point baseline, with 30 basis point stress slippage, were applied. The development of the rule took place from 2023 to 2025 after the last 3-hour gap. The thresholds used were 5, 10, and 15. A holdout period from January 1, 2026, to September 7, 2026, was used to test the rule. The results showed a return percentage of -42.83 and a drawdown percentage of 44.86. The win rate was 27.06 percent, with an average win percentage of 0.55 and an average loss percentage of 1.10. The expectancy percentage was -0.65. The stress test results showed a return percentage of -59.31 and a drawdown percentage of 59.68. The win rate was 12.94 percent, with an average win percentage of 0.58 and an average loss percentage of 1.29. The expectancy percentage was -1.05. The development results showed different return percentages and drawdown percentages for different thresholds and slippages. A bootstrap test was performed with 85 observations, a block length of 5, and 20,000 simulations. The expectancy percentage was -0.65, with a mean confidence interval of -0.86 to -0.41. The probability of a positive mean was 0. The premises of the test included a negative net expectancy, insufficient daily data, and untested actual 15-minute stop, target, and profit management. The conclusion was to reject the rule, with a size of zero. The results were based on data from https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, and were tested using the code test_rule.py, results_1Hour.json, and audit.py.
A news scan found the headline about a Frozen875 rule. This rule is based on a Simple Moving Average, which is a way to smooth out price changes, and an Exponential Weighted Moving Average Relative Strength Index, which is a measure of how fast the price is changing. The rule is: if the prior close is greater than the Simple Moving Average 100 and the Exponential Weighted Moving Average Relative Strength Index 2 is less than 10, then enter a long position, but exit if the Exponential Weighted Moving Average Relative Strength Index 2 is greater than 70 or after 10 bars. This rule was tested on 100 observations per frequency with a 25 basis point fee per side and a 10 basis point baseline plus 30 basis point stress slippage. The development of this rule took place from 2023 to 2025 after the last 3 hour gap. The thresholds used were 5, 10, and 15, and the rule was tested on a holdout period from January 1, 2026, to September 7, 2026. The results of the holdout period showed a return of -41.79% and a drawdown of 44.50%. The stress test showed a return of -61.68% and a drawdown of 62.94%. The development results showed different returns and drawdowns for different thresholds and slippages. A bootstrap simulation was also run with 20,000 simulations, which showed a mean expectancy of -0.51%. The premises of this test were that the net expectancy was negative, the daily N 0-2 was insufficient, and the actual 15 minute stop target profit management was untested. The conclusion was to reject the rule and size it to zero. The results include one open paper trade each for LINK/UNI returns, and the win rate excludes open trades. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The code used to test the rule is test_rule.py, results_1Hour.json, and audit.py.
A news scan found the headline 'LINK/USD hourly trend-pullback fails costed holdout'. The Frozen875 rule is a set of conditions to buy or sell a stock, in this case, LINK, which is a type of cryptocurrency. The rule is: if the previous close is greater than the Simple Moving Average of 100, which is a calculation to smooth out price data, and the Exponential Weighted Moving Average Relative Strength Index 2, a measure of how fast the price is changing, is less than 10, then buy the stock. The stock is sold when the Relative Strength Index 2 is greater than 70 or after 10 bars, which are units of time. This rule is used for long positions only, which means buying the stock with the expectation of selling it at a higher price. The analysis is based on 100 observations per frequency. The analysis also includes a fee of 25 basis points, which is 0.25%, per side, and a baseline of 10 basis points, and a stress slippage of 30 basis points, which is the difference between the expected price and the actual price. The development of this rule took place from 2023 to 2025 after the last gap of more than 3 hours in the data. The thresholds used were 5, 10, and 15. The holdout period, which is a set of data not used to develop the rule, was from January 1, 2026, to September 7, 2026. The results of the analysis are as follows: the return percentage was -45.89%, the drawdown percentage, which is the maximum loss, was 48.13%, the win rate percentage was 36.27%, and the average win percentage was 0.57%. The stress test, which is a simulation of the rule with a higher slippage, resulted in a return percentage of -64.09% and a drawdown percentage of 64.27%. The development results show different thresholds and slippages, with return percentages ranging from -58.09% to -99.94%. The bootstrap test, which is a simulation of the rule with random sampling, resulted in a mean expectancy percentage of -0.58% and a probability of a positive mean of 0. The premises of the analysis are that the net expectancy is negative, the daily data is insufficient, and the actual stop and target profit management is untested. The conclusion is to reject the rule, with a size of zero. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The code used is test_rule.py, results_1Hour.json, and audit.py.
A news scan found the headline 'LTC/USD hourly trend-pullback fails costed holdout'. The Frozen875 rule is a set of conditions used to make trading decisions, which includes prior close being greater than the Simple Moving Average of 100, and the Exponential Weighted Moving Average of the Relative Strength Index 2 being less than 10. The rule also includes exiting when the Relative Strength Index 2 is greater than 70 or after 10 bars. The next adjacent open is used for long positions only, with 100 observations per frequency. A 25 basis point fee per side, plus a 10 basis point baseline and 30 basis point stress slippage were applied. The development period was from 2023 to 2025, after the last gap of more than 3 hours. The thresholds used were 5, 10, and 15, with a fixed holdout period from January 1, 2026, to September 7, 2026. The quality of the data was evaluated, with 32296 original observations, 31426 used, 0 duplicates, 870 removed, and 4 missing. The data started on February 6, 2023, and ended on September 7, 2026. The holdout results showed a return percentage of -51.78, a drawdown percentage of 52.16, 92 trades, a win rate percentage of 14.13, an average win percentage of 0.50, and an average loss percentage of 0.998. The stress results showed a return percentage of -66.62, a drawdown percentage of 66.66, 92 trades, a win rate percentage of 5.43, an average win percentage of 0.63, and an average loss percentage of 1.29. The development results showed various thresholds and slippages, with returns ranging from -78.92 to -99.90. The bootstrap results showed a mean expectancy percentage of -0.79, with a 95% confidence interval of -1.00 to -0.58. The premises of the analysis were that the net expectancy was negative, daily N0-2 was insufficient, and the actual 15-minute stop/target/profit management was untested. The conclusion was to reject the strategy, with a size of zero. The data sources were https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The code used was test_rule.py, results_1Hour.json, and audit.py.
A news scan found the headline about a SOL/USD hourly trend-pullback that failed, costing a holdout. The Frozen875 rule is based on the prior close being greater than the Simple Moving Average of 100 and the Exponential Weighted Moving Average Relative Strength Index 2 being less than 10. This means the rule uses a type of average that gives more weight to recent prices and a measure of how fast the price is changing. The rule exits when the Relative Strength Index 2 is greater than 70 or after 10 bars. The rule is for long positions only, meaning it only buys, not sells. The rule was tested 100 times per frequency. The test included a 25 basis point fee per side and a 10 basis point baseline fee, with a 30 basis point stress slippage. This means the test included a small fee for each trade and a slightly larger fee to account for unexpected price changes. The development of the rule took place from 2023 to 2025, after the last gap of more than 3 hours in the market data. The rule was tested with thresholds of 5, 10, and 15, and was fixed from January 1, 2026, to September 7, 2026, using a chronological holdout, which means the rule was tested on data that had not been seen before, but was from the same time period as the training data. The quality of the data was checked, with 22,219 original observations, 17,809 used, 0 duplicates, 4,410 removed, and 7 missing. The data started on August 26, 2024, at 16:00:00 and ended on September 7, 2026, at 23:00:00. The holdout results showed a return of -51.66%, a drawdown of 52.82%, and a win rate of 25.42%. The stress results showed a return of -69.85%, a drawdown of 70.08%, and a win rate of 11.86%. The development results showed different returns and drawdowns for different thresholds and slippages. The bootstrap results showed a mean expectancy of -0.61% and a probability of a positive mean of 0. The premises of the test were that the net expectancy was negative, the daily N0-2 was insufficient, and the actual 15-minute stop/target/profit management was untested. The conclusion was to reject the rule, with a size of zero. The results also mentioned that the LINK/UNI returns included one open paper trade each and the win rate excluded open trades. The source of the data was https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The code used was test_rule.py, results_1Hour.json, and audit.py.
A frozen note, number 877, describes a trading strategy. The strategy was trained on data from 2016 to 2020, and then tested on new data from 2021 to 2025. The test period was also historically inspected. The strategy only considers complete trading periods, which are called cohorts, from 2026. The strategy buys when the first session open of the month is higher than the previous month's close and also higher than the 100-day moving average, which is a calculation of the average price over the last 100 days. The previous bar's average true range, or ATR, which is a measure of price volatility, is used to set a stop loss, which is an order to sell when the price falls to a certain level, and a target price, which is the desired sale price. The strategy sells half of the position when the price reaches one average true range above the buy price, and then moves the stop loss to the break-even point, which is the price at which the trade will not result in a loss. If the price moves in the opposite direction, the strategy sells the remaining half of the position. The strategy also uses a gap stop, which is a stop loss that is triggered when the price gaps, or jumps, above or below a certain level. All prices are adjusted for dividends, which are payments made to shareholders, and a 5 basis point fee, which is a small charge, and 20 basis points of slippage, which is the difference between the expected and actual prices of a trade. The strategy uses a diagnostic tool to test the effect of different position sizes, starting with an initial cash amount of 100. The strategy does not use leverage, which is borrowing money to invest. The results of the strategy are compared to a benchmark using a tool called vectorbt, which reconstructs the trades and agrees with the results to within a very small margin of 3.55e-13 percentage points. The drawdown, which is the peak-to-trough decline in the value of the portfolio, is calculated using synthetic daily fill and close prices, rather than actual 15-minute prices. The positions are normalized to full allocation, which means that the entire portfolio is invested, for comparison purposes. A bootstrap analysis, which is a statistical test, was run 10,000 times with a seed value of 90802, and the results showed a 95% mean interval of -0.768018% to -0.193765%. The strategy only uses historical data to size positions, and any failures during the training or holdout periods are kept at zero, even if the actual result was positive, as in the case of the 2026 counterexample. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the author's paper is https://pages.stern.nyu.edu/~afrazzin/pdf/Betting%20Against%20Beta%20-%20Frazzini%20and%20Pedersen.pdf. The strategy is not an exact leveraged betting against beta strategy. The study has 252 metrics observations for QLV90802, and is a deterministic study, which means that the results are not random, and is run using a local Python script called deterministicstudy.py.
A frozen note, number 877, was used for analysis. The analysis used data from 2016 to 2020 for training, and data from 2021 to 2025 was used as a holdout period, but was also historically inspected. The analysis only included completed cohorts from 2026. The strategy used was to buy at the first session open of the month if it was higher than the 100-day moving average of the previous month's close. The stop loss was set at 1 average true range, and the target was set at 1.5 average true ranges, with a time exit at the 10th session close. A moving average is a way to smooth out price data. The average true range is a measure of price volatility. The analysis did not include partial profits, and the gap stop open was used if there was ambiguity. All results were adjusted for daily systematic investment plan, with a 5 basis point fee and 20 basis points of slippage for each fill. A basis point is one hundredth of a percent. The analysis used fractional continuous sizing, with an initial cash amount of 100, one position, and no leverage. Leverage refers to using borrowed money to invest. The vectorbt exact fill reconstruction agreed with independent compounding within 3.80e-13 percentage points. The drawdown was calculated using synthetic per day fill and close slots, not the actual 15-minute path. Positions were normalized for full allocation for comparison, but not as a proposal. A bootstrap analysis was done with 10,000 iterations and a seed of 90802, resulting in a 95% mean interval of -0.879100% to -0.205256%. The historical profit and loss sizing was only used for training and holdout, and failures were kept at actual 0 even if there were positive counterexamples in 2026. The source of the data was https://runway.adityaperswal.workers.dev/data/stock-bars, and the author's paper was https://pages.stern.nyu.edu/~afrazzin/pdf/Betting%20Against%20Beta%20-%20Frazzini%20and%20Pedersen.pdf. The analysis was not an exact leveraged betting against beta strategy. The full 252 metrics observations for QLV90802 were used, and the study was deterministic, using the ledgers local in study.py.
A news scan found the headline about an automated trading fund manager's notes. The manager used a rule called Frozen875. This rule means that the manager looks at the price of Ethereum, or ETH, and its movement over time, using something called a Simple Moving Average, or SMA, and an Exponential Weighted Moving Average, or EWMA, of the Relative Strength Index, or RSI, which is a way to measure how fast the price is changing. The rule says to sell when the RSI is greater than 70 or after 10 time periods, and to buy when the RSI is less than 10 and the next open price is higher. The manager tested this rule using 100 observations per frequency, which is a way to look at how often something happens, and used a fee of 25 basis points, or 0.25%, per trade, plus a baseline fee of 10 basis points, or 0.10%, and a stress slippage of 30 basis points, or 0.30%, which is a way to measure the difference between the expected price and the actual price of a trade. The manager developed this rule from 2023 to 2025, after the last time the price had a gap of more than 3 hours, and used thresholds of 5, 10, and 15, which are like targets for the price to reach. The manager tested the rule from January 1, 2023, to September 7, 2026, using a holdout period, which is a way to test a rule without using the same data it was developed with. The results showed a return of -50.90% and a drawdown of 50.99%, which is a way to measure the biggest loss. The manager also tested the rule with different slippages, or differences between the expected price and the actual price, and found that the results were similar. The manager used a bootstrap method, which is a way to test a rule by repeating it many times, and found that the results were not positive. The manager concluded that the rule did not work well and should be rejected. The manager also noted that the results included one open trade each for LINK and UNI, and that the win rate did not include open trades. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees, and the code used to test the rule is in the files test_rule.py, results_1Hour.json, and audit.py.
A frozen note, number 877, contains information about a trading strategy. The strategy was trained using data from 2016 to 2020, and data from 2013 to 2015 is not available. The years 2021 to 2025 were used as a holdout period, but the market during this time was also historically inspected. The year 2026 was used to complete cohorts. The strategy buys at the open of the first session of the month if the previous month's close is greater than the 100-day moving average, which is a calculation of the average price of a stock over 100 days, used to determine trends. The strategy also uses a stop-loss, which is a price level at which to sell a stock to limit losses, set at one average true range, a measure of volatility, below the buy price. The target price is set at 1.5 average true ranges above the buy price, and the position is closed at the close of the 10th session. If the stock price reaches one average true range above the buy price, half of the position is sold, and the stop-loss is moved to the break-even price. The strategy also uses a gap stop, which is a stop-loss set at the open price if the stock gaps up or down, and the target price is capped. All prices are adjusted for dividends and splits, and a fee of 5 basis points, or 0.05%, and slippage of 5 basis points are applied to each fill, which is the execution of a trade. The strategy uses fractional continuous sizing, which means that the position size is adjusted based on the stock's volatility, and the initial cash is $100. The strategy does not use leverage, which is borrowing money to buy stocks. The vectorbt exact fill reconstruction agrees with independent compounding within 1.01e-13 percentage points, which means that the strategy's performance is accurate to a very small degree. The drawdown, which is the peak-to-trough decline in the stock's price, uses synthetic per-day fill/close slots, not the actual 15-minute path, and positions are normalized to full allocation for comparison, not as a proposal. A bootstrap analysis, which is a statistical method for estimating the uncertainty of a strategy's performance, was performed with 10,000 iterations and a seed of 90802, and the 95% mean interval is [-0.840583, 1.131843]%. The historical profit/loss and win/loss sizing are only used for training, and failures in training and holdout keep the actual profit/loss at zero, even if it is positive in 2026. The strategy's performance is compared to the EEMV, which is an exchange-traded fund that tracks the performance of emerging markets. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the author's paper is https://pages.stern.nyu.edu/~afrazzin/pdf/Betting%20Against%20Beta%20-%20Frazzini%20and%20Pedersen.pdf. The strategy is not an exact leveraged betting against beta strategy. The QLV90802 metrics have 252 observations, and the study is deterministic, meaning that it is based on historical data and does not involve random chance. The ledgers are local and can be found in the deterministic study.py file.