A quant starts with $245.69 in cash and $850.99 in account equity, with no current investments or pending orders. The goal is to achieve repeatable positive returns. A key constraint is having an edge, or advantage, that is measured and can be acted on, but not all available cash can be used. There are three premises: 1. A premium, or extra return, can make up for the risk of holding inventory, or investments, without being available to traders who pay the spread, or difference between buying and selling prices, twice a night. 2. A common diagnostic tool for exchange-traded funds, or ETFs, is flawed because it uses price instead of actual buying and selling data. 3. Previous research found that the premium decreased over time, and historical data shows that some investment products were closed, but this does not mean it is a universal rule. The first order effect is that more efficient matching of buyers and sellers can leave less inventory and result in smaller price concessions. The second order effect is that repeated buying and selling transfers more of the returns to intermediaries, such as fund managers, while management fees remain the same. The third order effect is that investors seeking higher returns may take on more risk, increasing the potential for large losses. These effects are hypotheses, not newly discovered causes. The strongest argument against this is that forced selling could restore the premium, but this would require new evidence of actual buying and selling data and net returns, not just higher volatility. The average gross return of 3.37 basis points per night for the SPY ETF is below the cost of a round trip, or buying and selling, of 4 basis points. This small difference can add up over 252 repetitions. The size of the investment remains zero. Sources include research papers from the New York Fed and a data website.
A study analyzed the relationship between volatility and overnight returns for the SPY stock. It found that in the early period, high volatility meant higher overnight returns, with a mean of 7.0960 basis points, compared to 1.7475 basis points in low volatility. However, in the later period, the means reversed, with high volatility having a mean of 2.9300 basis points and low volatility having a mean of 4.9050 basis points, based on 1,101 and 319 nights, respectively. The study also looked at returns above or below the prior 200-day Simple Moving Average, which is a measure of the stock's trend over time, and found means of 4.3619 basis points and -0.7598 basis points. Despite this, even the stronger returns were below the approximately 20 basis point round-trip base cost, which is the cost of buying and selling a stock. The study concluded that adding volatility, moving average, and price reversal into an entry score, which is a measure used to decide when to buy or sell a stock, is not effective. The study suggests that a new approach is needed, including new data on signed order flow, which refers to the direction of stock trades, and a fresh protocol. The data for this study came from https://runway.adityaperswal.workers.dev/data/stock-bars; statistics.json.
The feasibility of entering trades overnight in liquid ETFs is being considered. There are a few premises to consider: first, there are no current holdings or queued orders, so progress, target completion, and holding horizon are not applicable. Second, the closing prices on September 4 were SPY $770.19, QQQ $718.96, IWM $296.01, EFA $108.35, and EEM $68.70. With a position cap of $212.7475, it is possible to buy zero whole shares of the first three, one EFA, or three EEM. Third, SPY has a 50-day moving average of $756.8572 and a 200-day moving average of $712.1529, with an average true range of $5.29611 over the prior 20 sessions and an average daily range of $4.40956. An example stop-loss at $764.89389 and a target at $778.13417 are not validated for overnight exits, as the target is about 1.8 times the ordinary daily range, which is not a typical one-night expectation. Fourth, while liquidity is sufficient during regular trading hours, with a median daily dollar volume of $28.04 billion for SPY, this does not guarantee overnight spreads or the ability to exit fractional shares. The affordability of EFA and EEM does not address their negative holdouts. According to the source data from https://runway.adityaperswal.workers.dev/data/stock-bars and technicals.json, as well as Alpaca's market-data FAQ, daily bars do not accurately reconstruct overnight stop paths or guarantee official auction fills. The targets mentioned are illustrative, and no order is being proposed.
A trading fund manager made some simulations to decide how much to invest in the stock market. The manager looked at the performance of the SPY stock, which is a way to invest in the US stock market as a whole, over 1,420 nights. The manager found that the probability of making a profit was 37.1127%, the average profit was 0.3810%, and the average loss was 0.4890%. However, the average return was -0.1661% after considering the costs of buying and selling, which is known as a 10bp per side cost, meaning a 0.1% cost for each transaction. This means that the manager should not invest in this stock, according to a formula called the quarter-Kelly allocation, which helps decide how much to invest based on the potential risks and rewards. The manager also did some other simulations, including one with 10,000 samples of five trades, and another with 100,000 paths of 25 trades, to see how the investment would perform over time. The results showed that the investment would likely lose money, with a median profit and loss of -$8.51, and only 6.961% of the simulations resulting in a profit. The manager also used some statistical models, including a Gaussian and a Student-t model, to simulate the performance of the investment, and found that the average return was around -0.1752%. The manager used some data from a website and some scripts to run the simulations, and the results were stored in some files called simulations.json, barriers.json, and scripts.
A research paper, sr917, published in February 2020 and revised in August 2022, made a distinction between two types of exposure to the market, specifically costly unconditional exposure from 02:00-03:00 and a selective strategy based on negative closing-signed flow from 01:30-03:30. The paper used bid-ask data from 2004-2020. A follow-up report by the same author on July 1, 2026, found that the drift, or change, in the market from 02:00-03:00 was near zero from 2021-2025 and that the dispersion, or spread, of closing imbalances was compressed. This follow-up report is part of the same line of research and does not provide independent confirmation. Our own test of ETFs over the whole night used a different clock and instrument, and the positive gross return does not contradict the finding of a narrow hour fade in the original paper. Since futures data are unavailable and daily prices lack signed order flow, we do not claim to have exactly replicated the original paper. In conclusion, we reject using cash ETFs as proxies after considering costs and retain the original conditional mechanism as it has not been replicated. To conduct a new test, we need timestamped signed closing flow, executable quote history, and a fresh prospective window. Higher realized volatility alone cannot replace flow. No new fund is recommended without a numeric edge. The research papers can be found at https://www.newyorkfed.org/medialibrary/media/research/staff_reports/sr917.pdf and https://libertystreeteconomics.newyorkfed.org/2026/07/the-disappearing-overnight-drift/.
The fund's management fees are separate from the costs of buying and selling securities. On May 18, 2022, the fund's prospectus stated that it would invest in securities, ETFs, derivatives, and Treasury collateral to get close to the market's daily opening and closing prices, with a 0.55% annual management fee that does not include transaction or derivative costs. On July 14, 2023, the fund's board announced that the last trading day would be July 31 and the fund would likely be liquidated around August 10, not July 19. Before using old research as investment instructions, it's essential to check if the investment is still available and viable. The conclusion is that there is no entry for NSPY or NIWM based on the historical prospectus. The management fees go to the provider, transaction costs go to intermediaries, and the investor keeps the net return. A counterexample is that a fund's closure can be due to poor commercial scale rather than the absence of an anomaly, and the liquidation notice is not the sole cause. The fund's test using ETFs provides a separate rejection. The interest on cash collateral, futures basis, and roll mean ETF cash stock night test is not an exact replication of the fund's net asset value. The current product share valuation does not apply to historically liquidated funds. Sources include the SEC filings at https://www.sec.gov/Archives/edgar/data/1199046/000158064222002817/unifiedseriestrust_497.htm and https://www.sec.gov/Archives/edgar/data/1199046/000158064223003676/nightshares497s.htm.
A study analyzed five exchange-traded funds, or ETFs, with identical 2684-session histories from January 4, 2016, to September 4, 2026. The data, which includes open, high, low, and close prices, or OHLC, was split into two time periods: 2016-2020 and August 2021 to September 4, 2026. The SPY ETF had a gross mean overnight return of 3.373656 basis points, with a geometric breakeven point of 1.587993 basis points per side. The study found that the gross compound return of 56.98655% became -11.04239% at 2 basis points per side and -90.82762% at 10 basis points per side. A statistical analysis, called HAC 5 OLS, of the next day's open and close prices based on current and past intraday and 20-day volatility data showed a lagged beta of -0.015328, t-value of -0.620745, and R-squared of 0.02267. This means that price-only predictors did not identify signed flows. The strongest counterexample was the IWM ETF with a low cost of 15.96% and the SPY ETF with a lagged return of 6.38%, but their confidence intervals crossed zero and the base cost failed. The study also found that the overnight and intraday benchmarks exactly reconciled when corrected for passive close-to-close returns. The study's validity is limited by auction and dividend feed semantics, previously inspected dates, and bootstrap stationarity. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, with additional information in statistics.json and diagnostics.json.
A frozen protocol note, number 358, was made. This note looks at a time period from August 2021 to August 2026, which is the same market period that was previously inspected in other strategies. The return from open to close yesterday was negative, so the plan is to buy at today's close and sell at the following open. This signal was known before entering the trade, and it is based on a delayed price proxy, not an imbalance in signed orders. The first entry was made on January 5, 2021, and the last exit was made on August 27, 2026, with a total of 652 completed nights. There is a fee of 5 basis points and slippage of 5 basis points on each side of the trade. The vectorbt ledger, which is a long-only ledger, reconciles the independent net compounding within a very small margin of error, and it does not allow for negative cash. The drawdown includes the starting cash and fee marks, but it does not include unobserved intranight excursions. The mean payoff is -0.147999%, and the 95% mean confidence interval for a 5-trade block is -0.196250% to -0.103566%. The costs, including 0, 2, and 50 basis points, and raw price variants, are preserved in the archive, but they have not been tuned. The daily adjusted prices approximate the distributions and endpoints, but the market or limit venue cannot guarantee auctions, and there is no 15-minute exit or partial profit simulation in this backtest. This is a diagnostic adaptation, and it is not an exact replication of the 02-03 ET futures, nor is it a deployable stop or target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and it was generated using the study.py and diagnostics.py scripts.
A frozen protocol note, number 358, was used to analyze the SPY market from August 2021 to August 2026. This time period was previously examined in other strategies. The plan was to buy at the close of each session and sell at the following open. The first buy was on January 4, 2021, and the last sell was on August 31, 2026, with 1420 nights completed. There was a fee of 5 basis points and slippage of 5 basis points for each buy and sell. The results were checked using a long-only vectorbt ledger, which showed that the net compounding was accurate to within 1e-8 and that the cash balance was never negative. The average payoff was -0.166131%. A 95% confidence interval for the mean payoff over 5 trades was between -0.197787% and -0.135059%. Different cost scenarios, including 0, 2, and 50 basis points, were preserved in the archive but not adjusted for optimal performance. The prices used were adjusted for daily changes, but the simulation did not account for certain market factors, such as auctions or partial profits. This analysis is a diagnostic tool and does not exactly replicate the 02-03 ET futures or provide a deployable stop/target rule. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was conducted using study.py and diagnostics.py scripts.
A frozen protocol note, number 358, was used for training from 2016 to 2020. The strategy involved buying the SPY at the close of the market on days when the open-to-close return was negative, and then selling it at the next day's open. This signal was known before entering the trade, and it was based on a delayed price proxy, not an order imbalance. The first entry was on January 6, 2016, and the last exit was on December 31, 2020, with a total of 566 completed nights. The fee was 5 basis points, and slippage was also 5 basis points, on each side of the trade. The results were calculated using a long-only vector, and the net compounding was within 1e-8, with nonnegative cash. The drawdown included starting cash and fee marks, but did not include unobserved intranight excursions. The mean payoff was -0.173098%, with a 95% confidence interval of -0.266500% to -0.100093%. The costs were 0, 2, and 50 basis points, and raw price variants were preserved in the archive, but not tuned. Daily adjusted prices were used to approximate distributions and endpoints, but the market or limit venue cannot guarantee auctions, and there was no 15-minute exit or partial profit simulation in this backtest. This is a diagnostic adaptation, and not an exact replication of the 02-03 ET futures, and not a deployable stop or target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used to train a trading system from 2016 to 2020. The system was designed to buy at the close of each session and sell at the following open. It made its first entry on 2016-01-04 and its last exit on 2020-12-31, completing 1258 nights of trading. The system had a fee of 5 basis points and slippage of 5 basis points on each side of the trade. A long-only vectorbt ledger was used to record the buys and sells, and it showed that the system's net worth grew independently, with no negative cash balances, and was accurate to within 1e-8. The system's mean payoff was -0.157212%. A 95% confidence interval for the mean payoff over 5-trade blocks was -0.206346% to -0.112683%. The system preserved three variants of costs, 0, 2, and 50 basis points, and raw price variants, but did not tune them. The daily adjusted prices used in the system were approximate and did not guarantee exact market or limit venue auctions. The system did not simulate 15-minute exits or partial profits. This is a diagnostic adaptation and not an exact replication of the 02-03 ET futures or a deployable stop/target rule. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used to analyze the IWM market from August 2021 to August 2026, a period previously examined in other strategies. The analysis found that when the market had a negative return from open to close the previous day, it was a signal to buy at the close of the market today and sell at the open of the market the next day. This signal was known before entering the trade, and it was based on a delayed price proxy, not an order imbalance, which refers to a difference between the number of buy and sell orders. The first entry was made on January 4, 2021, and the last exit was on August 27, 2026, with a total of 711 completed trades at night. Each trade had a fee of 5 basis points, or 0.05%, and a slippage of 5 basis points, which is the difference between the expected price and the actual price. The analysis used a long-only strategy, meaning only buys were made, and the results were calculated using a vectorbt ledger, which is a tool used to track and analyze trades. The mean payoff was -0.139892%, and the 95% confidence interval for the mean payoff was between -0.196869% and -0.082145%. The analysis also preserved different variants of the data, including costs of 0, 2, and 50 basis points, but these were not used to adjust the results. The daily adjusted prices used in the analysis were approximate and did not take into account certain market factors, such as auctions. This analysis is a diagnostic tool and not a exact replication of a specific trading strategy, and it is not a deployable stop or target rule. The data used in the analysis came from https://runway.adityaperswal.workers.dev/data/stock-bars and was analyzed using the study.py and diagnostics.py scripts.
A frozen protocol note, number 358, was used to analyze the IWM overnight trading strategy from August 2021 to August 2026. This time period was previously examined in other strategies. The strategy involved buying at the close of each session and selling at the open of the next session. The first entry was on January 4, 2021, and the last exit was on August 31, 2026, with a total of 1420 completed nights. The fee was 5 basis points and slippage was 5 basis points on each side. The results were calculated using a long-only vectorbt staged buy/sell ledger, which reconciles independent net compounding within a small margin of error, and includes nonnegative cash. The mean payoff was -0.145874%. A 95% confidence interval for the mean payoff over 5-trade blocks was -0.189569% to -0.102042%. The backtest preserved variants with costs of 0, 2, and 50 basis points, as well as raw price variants, but these were not tuned. Daily adjusted prices were used to approximate distributions and endpoints. Note that this backtest did not simulate auctions, 15-minute exits, or partial profit taking, and is a diagnostic adaptation, not an exact replication of 02-03 ET futures or a deployable stop/target rule. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was conducted using study.py and diagnostics.py.
A frozen protocol note, number 358, was used for training from 2016 to 2020. The strategy involved buying at the close and selling at the next open when the previous day's return was negative. This signal was known before entering the trade, but the price used was delayed and did not account for order imbalance. The first entry was on 2016-01-05 and the last exit was on 2020-12-31, with a total of 602 completed nights. The fee was 5 basis points and slippage was also 5 basis points on each side of the trade. The results were calculated using a long-only vectorbt ledger, which reconciles the net compounding within a small margin of error, and includes starting cash and fees but excludes unobserved price movements during the night. The mean payoff was -0.180225%, with a 95% confidence interval of -0.269987% to -0.105063% over 5-trade blocks. Different cost variants were preserved but not used for tuning. The daily adjusted prices used in the study approximate real market distributions and endpoints, but do not guarantee auction prices and do not include 15-minute exits or partial profit simulations. This study is a diagnostic adaptation and not an exact replication of futures trading from 02-03 ET, and it is not a deployable stop or target rule. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was conducted using study.py and diagnostics.py.
A frozen protocol note, number 358, was used to train a system from 2016 to 2020. The system was designed to buy at the close of each session and sell at the following open. It started on January 4, 2016, and ended on December 31, 2020, completing 1258 nights. The system had a fee of 5 basis points and slippage of 5 basis points on each buy and sell. A long-only strategy was used, meaning the system only bought and never sold short. The system's performance was calculated using a method called vectorbt, which is a way to backtest trading strategies. The results showed a mean payoff of -0.133879 percent. A 95 percent confidence interval, which is a way to measure the reliability of the results, showed a range of -0.186687 to -0.082900 percent. The system was tested with different cost scenarios, including 0, 2, and 50 basis points. The prices used were adjusted daily and were approximate. The system was not designed to exactly replicate real-time trading and did not include features like stop-loss or take-profit rules. The data used for the study came from https://runway.adityaperswal.workers.dev/data/stock-bars and was analyzed using two scripts called study.py and diagnostics.py.
A frozen protocol note, number 358, was used to analyze the EFA market from August 2021 to August 2026, a period previously examined in other strategies. The analysis found that when the market had a negative return from open to close the previous day, it was a signal to buy at the close of the market today and sell at the open of the market the next day. This signal was known before entering the trade and was based on a delayed price proxy, not an order imbalance that was not signed. The first entry into the market was on January 4, 2021, and the last exit was on August 31, 2026, with 631 completed nights. The fee was 5 basis points and the slippage was also 5 basis points for each side of the trade. The analysis used a long-only vectorbt staged buy and sell ledger, which reconciles independent net compounding within a very small margin of error, and does not allow for negative cash. The drawdown includes the starting cash and fee marks, but does not include unobserved intranight excursions. The mean payoff was -0.240984 percent, with a 95 percent confidence interval of -0.309603 percent to -0.175063 percent for a 5-trade block. The costs of 0, 2, and 50 basis points, as well as raw price variants, were preserved in the archive but not tuned. The daily adjusted prices approximate distributions and endpoints, but the market or limit venue cannot guarantee auctions, and there is no 15-minute exit or partial profit simulation in this backtest. This analysis is a diagnostic adaptation and not an exact replication of the 02-03 ET futures, and it is not a deployable stop or target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used to analyze the EFA market from August 2021 to August 2026, which is the same period previously inspected in other strategies. The plan was to buy at the close of each session and sell at the following open. The first buy was on January 4, 2021, and the last sell was on August 31, 2026, resulting in 1420 completed nights. The cost was 5 basis points for the fee and 5 basis points for slippage on each side, which is 10 basis points per side. A long-only strategy was used, which means only buying was allowed, and the results were checked for accuracy. The average payoff was -0.197311 percent, and the 95 percent confidence interval for the mean of 5-trade blocks was -0.238236 percent to -0.157933 percent. Different cost scenarios, including 0, 2, and 50 basis points, were preserved in the archive but not adjusted for optimal performance. The prices used were adjusted daily and approximate the actual distributions and endpoints. However, the market or limit venue cannot guarantee auctions, and this backtest did not simulate 15-minute exits or partial profits. This analysis is a diagnostic adaptation and not an exact replication of the 02-03 ET futures or a deployable stop/target rule. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used for training from 2016 to 2020. The strategy was to buy at the close of the day if the open-to-close return was negative, and sell at the following open. This signal was known before entering the trade, and it used a delayed price proxy, not an order imbalance. The first entry was on 2016-01-08 and the last exit was on 2020-12-31, with 559 completed nights. The fee was 5 basis points and slippage was 5 basis points on each side. A long-only vector was used to stage buy and sell transactions, and the ledger was reconciled to ensure independent net compounding within a small margin of error, with nonnegative cash. The drawdown included starting cash and fee marks, but did not include unobserved intranight excursions. The mean payoff was -0.216290%, with a 95% confidence interval of -0.318499% to -0.127472%. Different cost variants were preserved in the archive, but not tuned. Daily adjusted prices were used to approximate distributions and endpoints, but the market or limit venue cannot guarantee auctions, and there was no 15-minute exit or partial profit simulation in this backtest. This is a diagnostic adaptation, not an exact replication of the 02-03 ET futures, and not a deployable stop or target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used to train a trading system from 2016 to 2020. The system was set to buy at the close of each session and sell at the following open. It made its first trade on January 4, 2016, and its last trade on December 31, 2020, completing 1258 trades overnight. The system was charged a fee of 5 basis points and allowed for 5 basis points of slippage on each trade. The results were calculated using a long-only strategy, which means the system only bought and never sold short, and the calculations were checked to make sure they were accurate to within a very small margin of error, known as 1e-8. The system's average profit was -0.178825 percent, meaning it lost value. A 95 percent confidence interval, which is a way to measure the reliability of the results, showed that the average profit was likely between -0.231093 percent and -0.127730 percent. The system was tested with different costs, including 0, 2, and 50 basis points, and the results were saved. The prices used were adjusted daily and were approximate. The system did not simulate real-world market conditions, such as auctions or partial profits. This test was used to diagnose issues and is not a exact replica of a real-world trading system, specifically the 02-03 ET futures replication, and should not be used as a rule for stopping or targeting trades. The data came from https://runway.adityaperswal.workers.dev/data/stock-bars and was analyzed using study.py and diagnostics.py scripts.
A frozen protocol note, number 358, was used to analyze the EEM market from August 2021 to August 2026, which is the same period previously examined in other strategies. The analysis found that when the market had a negative return from open to close the previous day, the strategy was to buy at the close of the market today and sell at the open of the market the next day. This signal was known before entering the trade, and it was based on a delayed price proxy, not an order imbalance that was not signed. The first entry into the market was on January 4, 2021, and the last exit was on August 28, 2026, with a total of 653 completed nights. The fee was 5 basis points and the slippage was also 5 basis points for each side of the trade. The analysis used a long-only vectorbt staged buy and sell ledger, which reconciles the independent net compounding within a very small margin of error, and the cash was always nonnegative. The drawdown included the starting cash and fee marks, but did not include any unobserved intranight excursions. The mean payoff was -0.239197%, and the 95% mean confidence interval for a 5-trade block was -0.311146% to -0.167636%. The analysis preserved the costs of 0, 2, and 50 basis points, as well as the raw price variants, in the archive, but did not tune them. The daily adjusted prices were used to approximate the distributions and endpoints, but the market or limit venue cannot guarantee auctions, and there was no 15-minute exit or partial profit simulation in this backtest. This analysis is a diagnostic adaptation and is not an exact replication of the ETF futures from 02-03 ET, and it is not a deployable stop or target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used to analyze the EEM market from August 2021 to August 2026, which is the same period previously inspected in other strategies. The plan was to buy at the close of each session and sell at the following open. The first buy was on January 4, 2021, and the last sell was on August 31, 2026, resulting in 1420 completed nights. The fee was 5 basis points and slippage was 5 basis points for each buy and sell. The analysis used a long-only strategy and reconciled the buy and sell ledger to ensure accurate net compounding within a small margin of error, 1e-8, and nonnegative cash. The drawdown calculation included starting cash and fee marks but did not include unobserved price changes during the night. The mean payoff was -0.195460%. A 95% confidence interval for the mean payoff over 5-trade blocks was -0.247297% to -0.145154%. The analysis preserved three variants of costs, 0, 2, and 50 basis points, and raw price variants in the archive, but did not tune them. Daily adjusted prices were used to approximate distributions and endpoints. The market or limit venue cannot guarantee auctions, and this backtest did not simulate 15-minute exits or partial profits. This analysis is a diagnostic adaptation and not an exact replication of the 02-03 ET futures or a deployable stop/target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the analysis was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used for training from 2016 to 2020. The strategy involved buying at the close of a day when the open-to-close return was negative and selling at the following open. This signal was known before entering the trade, and it was based on a delayed price proxy, not an order imbalance that was not signed. The first entry was on 2016-01-05 and the last exit was on 2020-12-30, with a total of 578 completed nights. A fee of 5 basis points and slippage of 5 basis points were applied to each side of the trade. The results were calculated using a long-only vector, with a buy and sell ledger that was reconciled independently to ensure non-negative cash and net compounding within a small margin of error, 1e-8. The mean payoff was -0.159961%, with a 95% confidence interval of -0.246600% to -0.074461%. Different cost variants, 0, 2, and 50 basis points, and raw price variants were preserved in the archive but not tuned. Daily adjusted prices were used to approximate distributions and endpoints, but the market or limit venue cannot guarantee auctions, and there was no 15-minute exit or partial profit simulation in this backtest. This is a diagnostic adaptation and not an exact replication of the 02-03 ET futures or a deployable stop/target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used for training from 2016 to 2020. The strategy was to buy at the close of each session and sell at the following open. The first trade was on January 4, 2016, and the last trade was on December 31, 2020, with a total of 1258 completed trades overnight. The fee was 5 basis points and slippage was 5 basis points for each buy and sell. The results were calculated using a long-only strategy, with a ledger that reconciles the buys and sells, and the net compounding was calculated independently, with a maximum error of 1e-8, and the cash was always nonnegative. The drawdown, or loss, includes the starting cash and fees, but does not include any potential losses that may have occurred during the night. The mean payoff, or average return, was -0.152876%. A 95% confidence interval for the mean payoff over 5 trades was between -0.210277% and -0.096504%. The costs were 0, 2, and 50 basis points, and the raw price variants were preserved in the archive, but not used to adjust the strategy. The daily adjusted prices were used to approximate the distributions and endpoints, but the market or limit venue cannot guarantee auctions, and there was no simulation of 15-minute exits or partial profits in this backtest. This is a diagnostic adaptation, not an exact replication of the ETF futures from 02-03 ET, and not a deployable stop or target rule. The source of the data is https://runway.adityaperswal.workers.dev/data/stock-bars, and the study was done using study.py and diagnostics.py.
A frozen protocol note, number 358, was used to analyze the QQQ market from August 2021 to August 2026, a period previously examined in other strategies. The analysis found that when the market had a negative return from open to close the previous day, it was a signal to buy at the close and sell at the following open. This signal was known before entering the trade and was based on a delayed price proxy, not an order imbalance, which refers to a difference in the number of buy and sell orders. The first entry was on January 5, 2021, and the last exit was on August 27, 2026, with 658 completed trades at night. The fee was 5 basis points, or 0.05%, and slippage, which is the difference between expected and actual trade prices, was also 5 basis points per trade. The analysis used a long-only strategy, meaning only buys were considered, and a vectorbt ledger, which is a tool used to track and analyze trades. The results showed a mean payoff of -0.136809%, and a 95% confidence interval of -0.197342% to -0.078657%. The analysis preserved different cost variants, but did not tune them. The daily adjusted prices used in the analysis were approximate and did not guarantee exact market or limit prices. This analysis is a diagnostic tool and not a replication of exact trading results or a deployable stop-loss rule. The data source was https://runway.adityaperswal.workers.dev/data/stock-bars, and the analysis used study.py and diagnostics.py scripts.
A frozen protocol note, number 358, was used to analyze the QQQ market from August 2021 to August 2026, which is the same period previously inspected in other strategies. The plan was to buy at the close of each session and sell at the following open. The first buy was on January 4, 2021, and the last sell was on August 31, 2026, resulting in 1420 completed overnight trades. The cost was 5 basis points for the fee and 5 basis points for slippage on each buy and sell. The analysis used a long-only strategy and ensured that the results were accurate to within a very small margin, 1e-8, and that the cash balance was never negative. The average profit per trade was -0.156518 percent, and the 95 percent confidence interval for the average profit over 5 trades was between -0.197996 percent and -0.115258 percent. Different cost scenarios, 0, 2, and 50 basis points, were preserved in the archive but not adjusted for optimal performance. The prices used were adjusted for daily changes and approximate the actual market conditions, but the analysis did not account for certain market factors like auctions and did not simulate exiting trades within 15 minutes or taking partial profits. This analysis is a diagnostic tool and not an exact replication of a specific trading strategy or a set of rules that can be used for actual trading. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars, and the analysis was done using the study.py and diagnostics.py programs.