Stats

A preregistered note, number 1253, was written before data was retrieved. The test used nine sector ETFs, which are a type of investment fund that tracks a specific sector, and ranked them based on their adjusted close returns over the prior five sessions at the weekly boundary. The worst performing ETF was bought at the next open if its own return was negative. The SPY, which is a ticker symbol for an ETF that tracks the S&P 500 stock market index, subtraction did not change the cross-sectional ranks. The prior 20 mean high-low range, which is a measure of price volatility, was used to set the initial stop, target, and half off levels. The test was trained on data from 2016 to 2020 and tested on data from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill and adjusted OHLC, which stands for open, high, low, and close prices. The full normalized capital was $10,000, with cash only and fees of 1 basis point, which is a unit of measurement for interest rates, and slippage of 15 basis points each side, including partials. The custom event cash and every equity point were verified by vectorbt from orders, with an exact final value of $6,939.16. The stop priority was set on daily collisions, with an adverse gap stop at open and target gap recognized. Partial and breakeven ambiguous days were assigned an adverse sequence. The continuous touch is optimistic versus 15-minute polling, even though the collision rule is conservative. The current-sized whole lots, half-rounding, and extended hour fills were not validated. This test is not an exact replication of the Nagel study, which used 48 industry and individual stocks, long-short, five daily sleeves, conditional beta hedge, closing liquidity supplier, and VIX dependence. The ETF cash adaptation tests a different payoff. The premises of this test are: recorded returns from a fixed rule, base positive edge must survive costs and uncertainty, and the whole family fails stress and confidence interval. The conclusion is that this is research-only and not for allocation. The sample size was 219, with a p-value of 0.547945205479452 and a mean return of -0.14778548536686914%. The years 2021 to 2025 had returns of -5.1554814620838885%, -4.253855505472382%, -5.28169377896509%, -12.922522067412046%, and -7.352503039406888%, respectively. The exit counts were 111 for stop, 43 for time, 38 for breakeven, 22 for target, and 5 for target gap. The trade sample counts were completed round trips, with partial legs aggregated. The return and drawdown are full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would worsen the net economics. All historical windows previously used by other fund studies were not pristine out-of-sample, and no winner was chosen from a sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076 and local research reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The test involved nine sector ETFs, which are a type of investment fund that tracks a specific sector. The ETFs were ranked based on their adjusted close returns over the prior five sessions at the weekly boundary. The worst performing ETF was bought at the next open if its own return was negative. The SPY, which is a type of ETF that tracks the S&P 500 stock market index, subtraction did not change the cross-sectional ranks. The prior 20 mean high-low range, also known as the average true range, was used to set the initial stop loss at 1R, the target at 2R, and the half off at 1R. The remaining stop entry was set at the maximum next weekly open. The test was trained on data from 2016 to 2020 and tested on data from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill and adjusted OHLC, which stands for open, high, low, and close prices. The survived ETF set was used. The test started with a full normalized capital of 10,000 and used cash only, with fees of 1 basis point and slippage of 5 basis points each side, including partials. A basis point is a unit of measurement for interest rates and fees. The custom event cash and every equity point were verified by vectorbt from orders. The exact final result was 10,752.96. The stop priority was set on daily collisions, and adverse gap stop at open, target gap recognized, and partial/BE ambiguous days were assigned adverse sequence. The continuous touch is optimistic versus 15-minute polling, even though the collision rule is conservative. The current-sized whole lots, half-rounding, and extended hour fills were not validated. The test was not an exact Nagel replication, which is a type of investment strategy. The original strategy used 48 industry and individual stocks, long-short, 5 daily sleeves, conditional beta hedge, closing liquidity supplier, and VIX dependence. The ETF cash adaptation tests different payoff. The premises of the test were: 1. recorded returns from fixed rule, 2. base positive edge must survive costs and uncertainty, and 3. whole family fails stress and CI. The conclusion of the test was research-only and no allocation. The sample size was 219, with a p-value of 0.5525 and a W-value of 1.549%. The mean return was 0.0521%, and the years 2021 to 2025 had returns of 3.569%, 4.553%, 3.225%, -4.912%, and 1.170%, respectively. The exit counts were 111 for stop, 43 for time, 38 for breakeven, 22 for target, and 5 for target gap. The trade sample counts were completed round trips, and partial legs were aggregated. The return and drawdown were full research capital and daily close marks, not account performance or worst intraday DD. No fixed monthly platform charges were deducted, which would have worsened the net economics. All historical windows previously used by other fund studies were not pristine OOS, and no winner was chosen from sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049 to 16076, and local research/reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The test involved nine sector ETFs, which are a type of investment fund that tracks a specific sector. The ETFs were ranked based on their adjusted close returns over the prior five sessions at the weekly boundary. The worst performing ETF was bought at the next open if its own return was negative. The SPY, which is a type of ETF that tracks the S&P 500 stock market index, subtraction did not change the cross-sectional ranks. The prior 20 mean high-low range, which is a measure of price volatility, was used to set the initial stop, target, and half off levels. The test was trained on data from 2016 to 2020 and tested on data from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill and adjusted OHLC, which stands for open, high, low, and close prices. The test survived the ETF set. The test started with a full normalized capital of $10,000, using cash only, with fees of 1 basis point and slippage of 15 basis points each side, including partials. A basis point is a unit of measurement for interest rates and fees. The custom event cash and every equity point were verified by vectorbt from orders. The exact final result was $7,258.72. The stop priority was set on daily collisions, with an adverse gap stop at open and target gap recognized. Partial and breakeven ambiguous days were assigned an adverse sequence. The test used continuous touch, which is a method of monitoring prices, and was optimistic versus 15-minute polling, even though the collision rule was conservative. The test did not validate current-sized whole lots, half-rounding, and extended hour fills. The test was not an exact replication of the Nagel study, which is a research paper on investment strategies. The premises of the test were: recorded returns from a fixed rule, base positive edge must survive costs and uncertainty, and the whole family fails stress and confidence interval. The conclusion of the test was research-only and no allocation. The sample size was 219, with a p-value of 0.547945205479452 and a mean return of -0.12724057828189275%. The returns for each year were: 2021, -9.575724080843905%; 2022, 1.4530726723529508%; 2023, -3.3536905856611954%; 2024, -13.306595788986852%; and 2025, -5.5637869206533015%. The exit counts were: stop, 104; target, 41; breakeven, 38; time, 29; and target gap, 7. The trade sample counts were completed round trips and partial legs aggregated. The return and drawdown were full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would have worsened the net economics. All historical windows previously used by other fund studies were used, not pristine out-of-sample data. No winner was chosen from a sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076 and local research using reversal.py.

A preregistered note, numbered 1253, was written before data was retrieved. The test involved nine sector ETFs, which are a type of investment fund that tracks a specific sector of the stock market. These ETFs were ranked based on their adjusted close returns over the prior three sessions at the weekly boundary. The worst performing ETF was bought at the next open if its own return was negative. The SPY, which is a type of ETF that tracks the S&P 500 stock market index, subtraction did not change the cross-sectional ranks. The prior 20 mean high-low range was used to set the initial stop at -1R, target at +2R, and half off at +1R, with the remaining stop entry at the maximum next weekly open. The test was trained on data from 2016 to 2020 and tested on data from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill and adjusted OHLC, which stands for open, high, low, and close, and survived ETF set. The test started with a full normalized capital of $10,000, using cash only, with fees of 1 basis point and slippage of 15 basis points each side, including partials. A basis point is a unit of measurement for interest rates and fees. The test used a custom event cash and every equity point verified by vectorbt from orders, with an exact final value of $5139.89. The test prioritized stops on daily collisions, with an adverse gap stop at open and target gap recognized, and partial or breakeven ambiguous days assigned adverse sequence. The test used continuous touch, which is a method of monitoring prices, and was optimistic versus 15-minute polling, even though the collision rule was conservative. The test did not validate current-sized whole lots, half-rounding, or extended hour fills. The test was not an exact replication of the Nagel method, which is a type of investment strategy, as it used ETF cash adaptation tests with different payoff. The premises of the test were: 1, recorded returns from a fixed rule; 2, base positive edge must survive costs and uncertainty; and 3, whole family fails stress and confidence interval. The conclusion of the test was research-only, with no allocation. The sample size was 225, with a p-value of 0.52 and a W-value of 1.350008917560452%. The mean return was -0.2750824997719447%, with years 2021, 2022, 2023, 2024, and 2025 having returns of -19.77584279508666%, -16.816823077388243%, 4.845641283599722%, -18.778632296479834%, and -9.553328538558715%, respectively. The exit counts were 123 for stop, 39 for time, 35 for breakeven, 24 for target, and 4 for target gap. The return and drawdown were based on full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would have worsened the net economics. All historical windows previously used by other fund studies were used, not pristine out-of-sample, and no winner was chosen from a sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076 and local research using reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The test involved nine sector ETFs, which are a type of investment fund that tracks a specific sector, and ranked them based on their adjusted close returns over the prior three sessions at the weekly boundary. The worst performing ETF was bought at the next open if its own return was negative. The SPY, which is a type of ETF that tracks the S&P 500 stock market index, subtraction did not change the cross-sectional ranks. The prior 20 mean high-low range, which is a measure of price volatility, was used to set the initial stop, target, and half off levels. The test was trained on data from 2016 to 2020 and tested on data from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill and adjusted OHLC, which stands for open, high, low, and close prices. The test survived the ETF set and used full normalized capital of $10,000, with cash only and fees of 1 basis point and slippage of 5 basis points each side, including partials. A basis point is a unit of measurement for interest rates and fees. The custom event cash and every equity point were verified by vectorbt from orders, with an exact final value of $8,060.95. The stop priority was set on daily collisions, with adverse gap stop at open, target gap recognized, and partial/breakeven ambiguous days assigned adverse sequence. The test used continuous touch, which is a method of monitoring prices, and was optimistic versus 15-minute polling, even though the collision rule was conservative. The test was not an exact replication of the Nagel study, which is a research paper on investment strategies, as it used ETF cash adaptation tests with different payoff. The premises of the test were: 1, recorded returns from fixed rule; 2, base positive edge must survive costs and uncertainty; and 3, whole family fails stress and confidence interval. The conclusion of the test was research-only, with no allocation. The sample size was 225, with a p-value of 0.52 and a W-value of 1.55%. The mean return was -0.075%, with a standard deviation of 1.839%. The returns for each year were: 2021, -12.57%; 2022, -9.16%; 2023, 14.72%; 2024, -10.95%; and 2025, -0.64%. The exit counts were: stop, 123; time, 39; breakeven, 35; target, 24; and target gap, 4. The trade sample counts were completed round trips, with partial legs aggregated. The return and drawdown were full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would have worsened the net economics. All historical windows previously used by other fund studies were used, not pristine out-of-sample, and no winner was chosen from a sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076 and local research/reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The test used nine sector ETFs, which are a type of investment fund that tracks a specific sector, and ranked them based on their adjusted close returns over the prior three sessions at the weekly boundary. The test bought the worst performing ETF at the next open if its own return was negative. The SPY, which is a type of ETF that tracks the S&P 500 stock market index, subtraction did not change the cross-sectional ranks. The prior 20 mean high-low range, which is a measure of volatility, was used to set the initial stop, target, and half off levels. The test used data from 2016 to 2020 for training and 2021 to 2025 for testing, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill and adjusted OHLC, which stands for open, high, low, and close prices. The test survived the ETF set and used full normalized capital of 10,000, with cash only and fees of 1 basis point, which is a unit of measurement for interest rates, and slippage of 15 basis points each side, including partials. The custom event cash and every equity point were verified by vectorbt from orders, with an exact final value of 5418.042992912734. The test used stop priority on daily collisions, adverse gap stop at open, target gap recognized, and partial/BE ambiguous days assigned adverse sequence. The test found that continuous touch is optimistic versus 15-minute polling, even though the collision rule is conservative. The test is not an exact Nagel replication, which refers to a specific type of investment strategy, as it uses ETF cash adaptation and tests a different payoff. The premises of the test are: 1, recorded returns from a fixed rule; 2, base positive edge must survive costs and uncertainty; and 3, the whole family fails stress and CI, which stands for confidence interval. The conclusion is research-only, with no allocation. The sample size is 225, with a p-value of 0.52 and a W-value of 1.3941282054574626%. The mean return is -0.25214047006550006%, with years 2021, 2022, 2023, 2024, and 2025 having returns of -21.661419014315452%, -14.01779110571465%, 6.137142136057849%, -18.073610246089235%, and -7.494630644957379%, respectively. The exit counts are 114 for stop, 39 for target, 35 for breakeven, 26 for time, and 11 for target gap. The trade sample counts completed round trips and partial legs aggregated. The return and drawdown are full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would worsen the net economics. All historical windows previously used by other fund studies were used, not pristine out-of-sample, and no winner was chosen from a sweep. The sources used include https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076 and local research/reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The test used nine sector ETFs, ranking them by their adjusted close returns over the prior three sessions at the weekly boundary. It bought the worst performing ETF at the next open if its own return was negative. Subtracting SPY, which is a type of stock market index fund, did not change the rankings. The initial stop loss was set at 1 times the prior 20-day mean high-low range, the target at 1.5 times this range, and half of the position was closed at 1 times this range. The test used data from 2016 to 2020 for training, 2021 to 2025 for testing, and 2026 for a retrospective extension. Late boundary-crossing trades were omitted. The test used all 10 series of 2685 aligned sessions from January 2016 to September 8, 2026, with no missing values and adjusted open, high, low, and close prices. The test started with a normalized capital of 10000, used cash only, and included fees of 1 basis point and slippage of 5 basis points on each side. The final capital was 8497.19. The test used a custom event to verify cash and equity points, and stop priority was used on daily collisions. The test was not an exact replication of the Nagel study, which used 48 industry and individual stocks, a long-short strategy, and a conditional beta hedge. The test used ETFs instead of individual stocks and did not include a VIX dependence. The premises of the test were to record returns from a fixed rule, to ensure a positive edge survived costs and uncertainty, and to test the whole family of strategies under stress and confidence intervals. The conclusion of the test was for research purposes only and did not recommend allocation. The sample size was 225, with a p-value of 0.52 and a W-value of 1.60. The mean return was -0.0524%, and the years 2021 to 2025 had returns of -14.63%, -6.11%, 16.13%, -10.18%, and 1.62%, respectively. The exit counts were 114 for stop losses, 39 for targets, 35 for breakeven, 26 for time, and 11 for target gaps. The return and drawdown were calculated using full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would have worsened the net economics. The historical windows used were previously used by other fund studies and were not pristine out-of-sample. No winner was chosen from a sweep of parameters. The sources used were a working paper from the National Bureau of Economic Research and a stock bars feed from a website, as well as local research and observations 16049-16076, and a Python script called reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The strategy involves nine sector ETFs, which are ranked based on their adjusted close returns over the prior five sessions at the weekly boundary. The worst performing ETF is bought at the next open if its return is negative. This is called a reversal strategy, meaning it goes against the recent trend. The SPY, a common stock market index, subtraction does not change the cross-sectional ranks, which means the comparison between the ETFs remains the same. The initial stop, target, and other parameters are set based on the prior 20 session mean high-low range, which is a measure of price volatility. The training period is from 2016 to 2020, the testing period is from 2021 to 2025, and there is a retrospective extension to 2026. Late boundary-crossing trades are omitted, which means trades that occur after the weekly boundary are not included. All ten series have 2685 aligned sessions from January 2016 to September 8, 2026, with no missing values, and the ETF set has survived. The normalized capital is $10,000, with cash only, and fees of 1 basis point and slippage of 5 basis points each side, including partials. A basis point is a unit of measurement for interest rates and fees. The custom event cash and every equity point have been verified by vectorbt from orders, with an exact final value of $11,078.92. The stop priority is on daily collisions, with an adverse gap stop at open, target gap recognized, and partial or breakeven ambiguous days assigned adverse sequence. Continuous touch is optimistic versus 15-minute polling, even though the collision rule is conservative. The premises of the research are: recorded returns from a fixed rule, a base positive edge must survive costs and uncertainty, and the whole family fails stress and confidence interval. The conclusion is that this is research-only, with no allocation. The sample size is 35, with a p-value of 0.6857, a win rate of 1.3326%, a loss rate of 1.9266%, and a mean return of 0.3083%. The years of return are listed, with 2026 having a return of 10.7892%. The exit counts are 14 for stop, 9 for breakeven, 8 for target, 3 for time, and 1 for target gap. The trade sample counts are completed round trips, with partial legs aggregated. The return and drawdown are for full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges are deducted, which would worsen the net economics. All historical windows previously used by other fund studies are not pristine out-of-sample, and no winner is chosen from a sweep. The sources of the data are listed, including a working paper from the National Bureau of Economic Research and a stock bars dataset. The observations are from 16049 to 16076, and the local research is from reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The strategy involves nine sector ETFs, which are ranked based on their adjusted close returns over the prior five sessions at the weekly boundary. The worst performing ETF is bought at the next open if its return is negative, a process known as a reversal strategy, which is a trading approach that involves buying assets that have performed poorly and selling those that have performed well. The SPY subtraction, which refers to the removal of the impact of the SPY ETF, does not change the cross-sectional ranks, meaning it does not affect the relative performance of the ETFs. The parent exits at the next weekly open without protective exits, which are trades designed to limit losses, and is not eligible for venue, a term that refers to the location or platform where trades are executed. The data used for training the strategy spans from 2016 to 2020, and the testing period is from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades, which occur when a trade is executed after the designated boundary, are omitted. The strategy was tested on all 10 series, with 2685 aligned sessions from January 2016 to September 8, 2026, with no missing values filled, and using adjusted OHLC, or open, high, low, and close prices. The initial capital was $10,000, with cash only, and fees of 1 basis point, or 0.01%, and slippage of 5 basis points, or 0.05%, each side, including partials. The final value was $13,996.98, verified by vectorbt from orders. The strategy prioritizes stop-loss orders on daily collisions, which occur when multiple trades are executed at the same time, and uses an adverse gap stop at open, which is a stop-loss order that is triggered when the price gaps below a certain level. The target gap is recognized, and partial or breakeven ambiguous days are assigned an adverse sequence. The strategy uses continuous touch, which is an optimistic approach, versus 15-minute polling, which is a more conservative approach. The results show a sample size of 35, with a p-value of 0.6857142857142857, and a W-score of 2.19947912344991%, and an L-score of 1.6359066382048224%, and a mean return of 0.9940721697869942%. The year 2026 had a return of 39.96983495286044%. The exit counts were 35, and the trade sample counts were completed round trips, with partial legs aggregated. The return and drawdown are based on full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would worsen the net economics. All historical windows previously used by other fund studies were used, and not pristine out-of-sample, and no winner was chosen from a sweep. The sources used include https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076, and local research reversal.py.

A preregistered note, numbered 1253, was written before data was retrieved. The test involved nine sector ETFs, which were ranked based on their adjusted close returns over the prior five sessions at the weekly boundary. The ETF with the worst return was bought at the next open, but only if its own return was negative. Subtracting the return of the SPY ETF did not change the cross-sectional ranks. The prior 20-day mean high-low range was used to set the initial stop loss at 1R, the target at 1.5R, and the half off at 1R, with the remaining stop entry set at the maximum next weekly open. The test was trained on data from 2016 to 2020 and tested on data from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill and adjusted OHLC. The full normalized capital was $10,000, with cash only and fees of 1 basis point and slippage of 5 basis points on each side, including partials. Custom events, including cash and every equity point, were verified by vectorbt from orders, with an exact final value of $11,248.15. The stop priority was set on daily collisions, with an adverse gap stop at open and target gap recognized. Partial and breakeven ambiguous days were assigned an adverse sequence. The test used continuous touch, which is optimistic versus 15-minute polling, even though the collision rule is conservative. The test did not exactly replicate the Nagel study, which used 48 industry and individual stocks, long-short, 5 daily sleeves, conditional beta hedge, closing liquidity supplier, and VIX dependence. The ETF cash adaptation tested a different payoff. The premises of the test were: 1, recorded returns from a fixed rule; 2, base positive edge must survive costs and uncertainty; and 3, the whole family fails stress and CI. The conclusion is that this is research-only and not for allocation. The sample size was 219, with a p-value of 0.5525, a W-value of 1.5862%, and an L-value of 1.7960%. The mean return was 0.0727%. The returns by year were: 2021, -1.2578%; 2022, 10.7856%; 2023, 5.3258%; 2024, -5.3318%; and 2025, 3.1232%. The exit counts were: stop, 104; target, 41; breakeven, 38; time, 29; and target gap, 7. The trade sample counts were completed round trips, with partial legs aggregated. The return and drawdown are full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would worsen the net economics. All historical windows previously used by other fund studies were not pristine out-of-sample, and no winner was chosen from a sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076 and local research using reversal.py.

A preregistered note, number 1253, was written before data was retrieved. The test used nine sector ETFs, which are a type of investment fund that tracks a specific sector, and ranked them based on their adjusted close returns over the prior five sessions at the weekly boundary. The worst performing ETF was bought at the next open if its own return was negative. The SPY, which is a type of ETF that tracks the S&P 500 stock market index, subtraction did not change the cross-sectional ranks. The parent exits at the next weekly open without protective exits, and is not eligible for a specific trading venue. The test was trained on data from 2016 to 2020, tested on data from 2021 to 2025, and retrospectively extended to 2026, with late boundary-crossing trades omitted. The test used all ten series and 2685 aligned sessions from January 2016 to September 8, 2026, with no missing value fill, adjusted OHLC, which stands for Open, High, Low, Close, and survived ETF set. The test started with a full normalized capital of 10,000, using cash only, with fees of 1 basis point and slippage of 5 basis points each side, including partials. A basis point is a unit of measurement for interest rates and fees. The test was verified by vectorbt from orders, with an exact final value of 10,550.62. The test had a stop priority on daily collisions, with an adverse gap stop at open, and a target gap recognized, with partial and break-even ambiguous days assigned an adverse sequence. The test used continuous touch, which is a method of monitoring prices, and was optimistic versus 15-minute polling, even though the collision rule was conservative. The test did not exactly replicate the Nagel replication, which is a type of investment strategy, as it used ETF cash adaptation tests with different payoffs. The premises of the test were: recorded returns from a fixed rule, a base positive edge must survive costs and uncertainty, and the whole family fails stress and confidence interval. The conclusion of the test was research-only, with no allocation. The sample size was 219, with a p-value of 0.5479, a W-value of 2.3086%, and an L-value of 2.6229%, with a mean of 0.0793%. The test had annual returns of 17.56% in 2021, -17.78% in 2022, 12.30% in 2023, 2.56% in 2024, and -5.22% in 2025. The test had 219 exit counts, with trade sample counts completed round trips and partial legs aggregated. The return and drawdown were based on full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would have worsened the net economics. All historical windows previously used by other fund studies were used, and not pristine out-of-sample, with no winner chosen from a sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076, and local research/reversal.py.

A preregistered note, numbered 1253, was written before data was retrieved. The strategy involves nine sector ETFs, ranking them by their adjusted close returns over the prior five sessions at the weekly boundary. The worst performer is bought at the next open if its return is negative. Subtracting SPY, which is a type of stock market index fund, does not change the rankings. The initial stop loss is set at one times the prior 20 session mean high-low range, the target is set at one and a half times this range, and half of the position is closed at one times this range. The remaining position is stopped out at the next weekly open. This strategy was tested on data from 2016 to 2020, and then applied to data from 2021 to 2025, with a retrospective extension to 2026. Late boundary-crossing trades were omitted. The test used all 10 series of 2685 aligned sessions from January 2016 to September 8, 2026, with no missing values filled, and adjusted open, high, low, and close prices. The test started with a normalized capital of 10000, using cash only, with fees of 1 basis point and slippage of 5 basis points on each side, including partials. A basis point is a unit of measurement for interest rates and fees. The results were verified by vectorbt, a backtesting framework, and the final value was 9321.38. The strategy prioritizes stop losses on daily collisions, and uses an adverse gap stop at the open, with target gap recognition. The results show a sample size of 219, with a p-value of 0.5297, a win rate of 1.2919%, and a loss rate of 1.4892%, with a mean return of -0.0161%. The returns by year were -6.4886% in 2016, -11.2381% in 2017, -4.0142% in 2018, -0.5455% in 2019, and 17.6408% in 2020. The exit counts were 103 for stop losses, 38 for targets, 37 for time stops, 37 for breakeven stops, and 4 for target gaps. The return and drawdown are based on the full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would have worsened the net economics. The historical windows used were previously used by other fund studies, and not pristine out-of-sample. No winner was chosen from a sweep. The sources used include a research paper from the National Bureau of Economic Research and a stock bars dataset from a website. The local research script used was reversal.py, with observations from 16049 to 16076.

A preregistered note, number 1253, was written before data was retrieved. This note is about a trading strategy that uses nine sector ETFs. The strategy ranks these ETFs based on their adjusted close returns over the prior five sessions at the weekly boundary. It buys the worst performing ETF if its return is negative. The strategy does not use protective exits and is not eligible for certain trading venues. The strategy was tested on data from 2016 to 2020 and then from 2021 to 2025, with a retrospective extension to 2026. Late trades that crossed boundaries were omitted. The strategy used all ten series of data with 2685 aligned sessions from January 2016 to September 8, 2026, with no missing values. The strategy started with a normalized capital of $10,000, using cash only, with fees of 1 basis point and slippage of 5 basis points per side. A basis point is a unit of measurement, equal to 0.01%. The strategy's performance was verified by vectorbt from orders, with an exact final value of $9,339.54. The strategy used a stop priority on daily collisions, with an adverse gap stop at open and a target gap recognized. Partial and break-even days were assigned an adverse sequence. The strategy used continuous touch, which is a method of monitoring prices, and was optimistic versus 15-minute polling. The strategy was not an exact replication of the Nagel strategy, which used 48 industry and individual stocks, long and short positions, and a conditional beta hedge. The ETF cash adaptation tested a different payoff. The premises of the strategy were: one, recorded returns from a fixed rule; two, a base positive edge must survive costs and uncertainty; and three, the whole family fails stress and confidence interval tests. The conclusion is that this is research only and not for allocation. The sample size was 219, with a p-value of 0.54337899543379 and a mean return of 0.031509207882245496%. The returns for each year were: 2016, 7.380320007257146%; 2017, -5.6854951288370925%; 2018, 6.974500634784531%; 2019, 21.73115524058902%; and 2020, -29.182516886926045%. The exit counts were 219. The trade sample counts were completed round trips, with partial legs aggregated. The return and drawdown are for full research capital and daily close marks, not account performance or worst intraday drawdown. No fixed monthly platform charges were deducted, which would worsen the net economics. All historical windows were previously used by other fund studies, not pristine out-of-sample. No winner was chosen from a sweep. The sources used were https://www.nber.org/system/files/working_papers/w17653/w17653.pdf and https://runway.adityaperswal.workers.dev/data/stock-bars?feed=sip&adjustment=all, with observations 16049-16076 and local research using reversal.py.

A news scan found the headline about HYPE 1Hour frozen trend-pullback. The costed diagnostic loses 60.18%. A frozen note was made before history download. The rule was that the prior close had to be greater than the Simple Moving Average of 100, which is a calculation to smooth out price data, and the RSI2 EWMA alpha 0.5 adjust false had to be less than 10. The next adjacent open was then considered, and the exit was prior to the RSI, or Relative Strength Index, which measures the speed and change of price movements, being greater than 70 or after 10 bars. The development period was from 2025 to 2026, and the holdout period was from January 1, 2026, to September 8, 2026. Each period started with no existing positions, or flat. The development thresholds of 5, 10, and 15 times the baseline and stress could not be calculated because the history started on February 17, 2026. The Vectorbt system was used with fees of 25 basis points, or 0.25%, and a slip of 30 basis points, or 0.30%, with a baseline of 60 basis points, or 0.60%, and a stress of the same amount. The fee approximation is not exact and is based on credited-asset accounting, with no 15-minute stop, target, or profit management simulated. The quality of the data was as follows: the time frame was 1 hour, there were 24 pages, and 4882 data points, with the first point on February 17, 2026, at 13:00:00 and the last point on September 8, 2026, at 23:00:00, with one missing point and 3805 zero points. The window was the holdout period, the threshold was 10, the slip was 0.003, and the return percentage was -60.18%, with a drawdown percentage of 60.19%, 89 data points, a win rate percentage of 29.21%, an average win percentage of 0.79%, an average loss percentage of 1.76%, and an expectancy percentage of -1.01%. The stress return was -76.66%, with 89 data points, a mean win of 0.79%, a mean loss of 1.98%, and an expectancy of -1.61%. There are three premises: one, no training means no qualified model; two, the current top 10 membership creates survivorship, and reused 2026 data is not pristine unseen data; three, zero-volume bars do not establish executable liquidity. The hourly data had one missing interval while flat, which retains warmup continuity limitation, with no interpolation, and a held gap would invalidate the data. The conclusion is to not move the split date to manufacture training, with a size of 0. A strong counterexample is the daily data, which had a return of 18.4%, but this is based on only four outcomes and cannot rescue the hourly failure. The source of the data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The results can be reproduced using the hype.py code, with detailed results observation 16041 and 16042.

A news scan found the headline about HYPE 1Day frozen trend-pullback. The rule for this trading strategy is that the prior close must be greater than the Simple Moving Average of 100, which is a calculation to smooth out price data, and the RSI2 EWMA alpha 0.5 adjust false must be less than 10. The next step is to look at the adjacent open, and exit when the RSI, or Relative Strength Index, which measures the speed and change of price movements, is greater than 70 or after 10 bars. This strategy was developed in 2025 and tested from January 1, 2026, to September 8, 2026, starting with no balance. The thresholds of 5, 10, and 15 times the baseline and stress levels could not be calculated because the history started on February 17, 2026. The Vectorbt system was used with fees of 25 basis points, which is a unit of measurement for interest rates and fees, and a slip of 30 basis points, which is the difference between the expected and actual price of a trade. The results showed a return of 18.3966%, a drawdown of 10.6242%, and a win rate of 75%. The average win was 6.8249% and the average loss was 2.6620%. The expectancy, which is the average profit or loss of a trade, was 4.4531%. The stress return was 15.5889%, with 4 trades, a win rate of 75%, an average win of 6.1843%, and an average loss of 3.2458%. The premises of this strategy are that no training means no qualified model, the current top 10 membership creates survivorship bias, which is the error of only considering the winners, and reused 2026 data is not pristine unseen data. The conclusion is that the sample size is too small, with only 4 trades, and the results cannot be used to make decisions. The source of this data is https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees. The results can be reproduced using the hype.py code, with detailed results observation 16041 and 16042.

A news scan found the headline 'Midnight CLI breakout diagnostic: five gross winners fail execution qualification'. The automated trading fund manager ran a test called CLI with a strategy called breakout, looking back 20 periods and checking the last 365 days across 7 major symbols. The results showed that for each symbol, the return, drawdown, win rate, and number of observations were saved. For example, BTC had a gross return of 5.929854%, a drawdown of 9.036712%, a win rate of 40%, and 20 observations. SOL and XRP were counterexamples to the idea that the market is bearish, with returns of 31.42404% and 55.13033% respectively. The manager then inspected the code in engine.js and bars.js and found that the signal to close a trade is based on the same close price, without considering fees or slippage, and without holding out any data. The data may include incomplete days, with up to 366 bars. However, these results should not be confused with actual future payoffs. The manager also reviewed independent records of daily and hourly data, which showed that none of the training data qualified for execution, with selected symbols such as UNI having a return of -42.91764%, a drawdown of 72.21537%, and 61 observations. The mean win was 5.26321% and the mean loss was 6.76089%. SOL was not considered due to missing exit data. The manager concluded that the CLI should not be used as a basis for deployment due to soundness limits, global reuse in 2026, survivorship, missing bars, and 15-minute exit resolution. The data sources are https://runway.adityaperswal.workers.dev/data/crypto-bars and https://docs.alpaca.markets/us/docs/crypto-fees.

A diagnostic test was run on the stock AVO using a momentum strategy with a lookback period of 20 days and 730 days of data, which included 500 bars and 24 trades. The test had two premises: first, that generic momentum does not estimate a dated earnings edge, and second, that the Command Line Interface (CLI) return was negative with no independent CLI holdout. The conclusion of the test was negative, based on a cross-check, and separate chronological vector backtest events were recorded. The data used for the test can be found at https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=AVO. The results of the test should not be used to determine the probability of a run-up in performance using a buy-holdout strategy.

A preregistered prior-session run-up strategy for AVO was tested using primary earnings data from 67 SEC 8K filings from January 2021 to June 2026. The data from December 19, 2024, was excluded due to a preannouncement on November 5. The strategy was trained on data from 2021 to 2024 and tested on data from 2025 to 2026. The strategy used a vectorbt long-only approach with a 10 basis point fee and 10 basis points of slippage on each side. The results showed that the baseline strategy had a return of -15.6021% in training and -7.3318% in holdout, with a maximum drawdown of 9.372%. A conservative stop-loss strategy using 1 ATR stop and 1.5 ATR target had a return of -20.4866% in training and -4.01016% in holdout. All 16 holdout variants of the strategy lost money. A news scan found the headline that the first release of AMC was not verified historically, so it is not a valid tradable strategy. The conclusion is to reject the strategy, as it does not have a necessary edge and has soundness limits due to missing archival advance notices, a short selected cohort, and adjusted daily OHLC not having 15-minute fills. The market data used can be found at https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=AVO&timeframe=1Day&feed=sip&adjustment=all. The research was reproducible using a 100k bootstrap seed and the research/test.py file.

A computer program executed a test on the stock MATX using a breakout strategy. The test looked back 20 days over a period of 1700 days, with a stop loss of 3% and a target gain of 4.5%. The test resulted in 1165 trades, with a return of -30.82459%, a maximum drawdown of 41.17376%, and a win rate of 28.86598%. The average trade return was -0.35229%, compared to a buy and hold return of 158.23066%. The test used data from https://runway.adityaperswal.workers.dev/data/stock-bars. This test is different from a previous test that used a next-open rule. The results of this test should not be directly compared to the previous test, and the difference in results should not be attributed solely to the data feed used. The test was run as a diagnostic and the output should be preserved. The test did not hold out any data, but a separate test will be run with holdout data for 2026.

A preregistered note, number 1013, analyzed the stock MATX using a consolidated SIP adjustment on daily bars from 2021 to September 4, 2026, which can be found at https://runway.adityaperswal.workers.dev/data/stock-bars?symbols=MATX&timeframe=1Day&feed=sip&adjustment=all. The analysis was divided into training from 2022 to 2024, validation in 2025, and a holdout period from January 1 to September 4, 2026, with a reset to flat. The strategy was to go long on the next open if the close was greater than the previous 20-day high and the 50-day simple moving average, which is a calculation of the average price of a stock over a 50-day period. The stop loss was set at 1 times the average daily range of the previous 20 days, and the target was set at 1.5 times this range. The maximum number of sessions was 10. Entry day exits were included in two daily execution slots, with a gap stop paid at the open and a gap target capped at the target, both of which were touched intraday, and no same-day reentry was allowed. Fees were 5 basis points plus slippage of 10 basis points on each side, with a stress test of 25 basis points slippage. The results were reconciled using vectorbt, a backtesting library, which is a tool used to test trading strategies, and an independent ledger, with a daily marked drawdown, which is the maximum loss from a peak to a trough, not 15-minute venue execution. The training results showed a return of 12.7051%, a drawdown of 13.9755%, 38 trades, a win rate of 52.6316%, and an average win of 3.4027 and average loss of 2.9987. The validation results showed a return of 6.3149%, a drawdown of 2.977%, and 5 trades. The holdout results showed a return of -1.1617%, a drawdown of 10.7850%, 13 trades, a win rate of 46.1538%, and an average win of 3.4881 and average loss of 3.0562. The stress test results showed a return of -4.9422% and a drawdown of 13.6811%. A fixed baseline was not changed after training, with only a 9-neighbor sweep, which is a method of analyzing nearby data points. A correction to the code explicitly prioritized the gapped target before the later intraday stop, but the results were unchanged. There is no claim of causality or execution at 15-minute observations. The selected current ticker, small holdout, and corporate action cash and executable overnight quote gaps mean that a positive result alone would not qualify. In this case, the point edge also fails, and the allocation is zero. A news scan found the headline related to MATX, but it is not included here as it is not part of the original note.

The QSEC90808 fund uses a fixed strategy, focusing on the top 1 positive sector momentum, with 9 ETFs. The strategy was trained on data from 2017 to 2020, held out from 2021 to 2025, and recently applied from 2026 through September 4. It manages risk with a 1 ATR stop and a 1.5 ATR target, with half of the position plus 1 ATR, then a breakeven point. The daily stop is the first ambiguous path, with no overlap, and the next open is lagged for the signal. There are 5 basis point fees and 20 basis points of lippage per fill. An event ledger, signed independently, reconciles the vectorbt with a maximum USD error of 3.637978807091713e-12. Fees are included on every partial fill. The strategy parameters are: stop at 4, same bar breakeven at 1, gap stop at 1, target at 1, gap target at 1, and boundary at 1. The data used is from 2016, with 2684 bars per symbol, and includes all adjusted total return proxy, cash dividends, exact fees, 15-minute quotes, and whole shares, but is not validated under the Pattern Day Trader rule. The out-of-sample data is chronological, but not pristine, due to previous market window reuse. The parent cannot authorize an order without required exits. This rule was not active on September 8 for the monthly entry. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, with notes frozen on 11/10/11 and 11/14, and strategy notes from 11/21.

The QSEC90808 primary managed stress holdout uses a fixed strategy based on the top 1 positive sector momentum, which is calculated from 9 ETFs. The strategy was trained on data from 2017 to 2020, and its performance is being tested on data from 2021 to 2025, with the most recent data going up to September 4, 2026. The strategy uses a 1 ATR stop and a 1.5 ATR target, with a goal of beating the market, and it has a daily stop on the first ambiguous path. The strategy also includes a non-overlap rule and uses the next open as a lagged signal. There are fees of 5 basis points and lippage of 20 basis points per fill. The event ledger is independently reconciled and has a maximum USD error of 3.637978807091713e-12. All fees are included, even on partial fills. The strategy has specific rules, including a stop at 27, a target at 13, and a best effort rule at 3. The data used is from 2016, with a floor of 2684 bars per symbol, and includes cash dividends, exact fees, 15-minute quotes, and whole-share data, but has not been validated. The out-of-sample data is chronological but not pristine due to previous market window reuse. The parent strategy cannot authorize an order without required exits, and this rule is not active on September 8 for monthly entry. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the notes were frozen on November 11, 2021, with a strategy date of November 21. The notes were also previously frozen on October 11 and October 14.

The QSEC90808 primary managed stress train uses a fixed strategy with a top 1 positive sector momentum, which is based on the performance of nine exchange-traded funds, or ETFs, over a specific period of time, known as a momentum indicator. The train period is from 2017 to 2020, the holdout period is from 2021 to 2025, and the recent period is from 2026 through September 4. The strategy is managed with a 1 ATR stop, which is a volatility-based stop-loss, and a 1.5 ATR target, where ATR stands for Average True Range, a measure of market volatility. The strategy also includes a half plus 1 ATR then break even, or BE, rule, which means that if the strategy's performance reaches a certain level, it will aim to break even. The daily stop is based on the first ambiguous path, and there is no overlap, with the next open lagged signal. The strategy includes fees of 5 basis points, or 0.05%, and 20 basis points, or 0.2%, of lippage per fill, which is the difference between the expected price of a trade and the actual price. A signed event ledger independently reconciles the vectorbt, a backtesting library, with a maximum USD error of 3.637978807091713e-12. Fees are included on every partial fill. The strategy parameters are: target is 9, stop is 23, same bar break even is 7, gap target is 7, and break even is 2. The data used is from 2016, with a floor of 2684 bars per symbol, and includes all adjusted total return proxy, cash dividends, exact fees, 15-minute quotes, whole-share, and PDT, or pattern day trader, rules, which are unvalidated. The out-of-sample, or OOS, period is chronological, but not pristine, due to previous market window reuse. The parent strategy cannot authorize an order without required exits. This rule is not active on September 8 for the monthly entry. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the notes were frozen on 10/11 and 10/14, with a strategy date of 11/21.

The QSEC90808 primary managed base recently used a fixed strategy with a monthly top 1 positive sector momentum, which included nine exchange-traded funds, or ETFs. The strategy was trained on data from 2017 to 2020, held out from 2021 to 2025, and recently used data from 2026 through September 4. The strategy managed a 1 ATR stop and a 1.5 ATR target, with half of the position and then a break-even, or BE, point, using a daily stop on the first ambiguous path, and a non-overlap with the next open lagged signal. The strategy included 5 basis point fees and 5 basis point slippage per fill. A signed event ledger independently reconciles the vectorbt with a maximum USD error of 1.8189894035458565e-12. Fees were included on every partial fill. The strategy parameters were set as follows: stop at 4, same bar BE at 1, gap stop at 1, target at 1, gap target at 1, and boundary at 1. The data used was from 2016, with a floor of 2684 bars per symbol, and included all adjusted total return proxy, cash dividends, exact fees, 15-minute quotes, whole shares, and PDT unvalidated. The out-of-sample, or OOS, data is chronological but not pristine due to previous market window reuse. The parent cannot authorize an order without required exits. This rule was not active on September 8 for the monthly entry. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, with frozen notes from November 11, 2021, and strategy from November 21, 2021.

The QSEC90808 fund uses a fixed strategy based on the top 1 positive sector momentum, which is calculated monthly and uses 9 ETFs. The strategy was trained using data from 2017 to 2020, and its performance is being tested on data from 2021 to 2025, with the most recent data going up to September 4, 2026. The fund is managed with a stop-loss of 1 ATR, a target of 1.5 ATR, and a trailing stop. ATR stands for Average True Range, which is a measure of market volatility. The fund also uses a daily stop and a signal that is delayed by one day. There is a fee of 5 basis points, which is 0.05%, and a slippage of 5 basis points per fill. The fund's performance is tracked independently using a ledger that records all events, and the maximum error is 3.637978807091713e-12 USD. The fund's data goes back to 2016 and includes 2684 bars per symbol, with all adjustments made for total return, including cash dividends and exact fees. The data is based on 15-minute quotes and whole shares, but has not been validated. The out-of-sample performance is chronological, but not pristine, due to the reuse of previous market windows. The parent fund cannot authorize an order without the required exits. This rule is not active for the monthly entry on September 8. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the notes were frozen on November 11, 2021, with a strategy update on November 21. The specific parameters for the strategy are: there are 27 stops, 13 targets, 7 same-bar break-evens, 5 gap targets, 4 gap stops, and 3 break-evens.