Stats

The model used 20,000 paths and 60 sessions, with historical open and previous close ratios, to analyze the stock PSIX. The model also considered fees of 1 basis point and a slip of 15 basis points per side. The results showed that 73.045% of the time, the stop was reached first, and 26.720% of the time, the target was reached first. The mean net result was -0.8900%, and the median was -10.2254%. A half-Kelly strategy was also applied, which resulted in a constrained stock weight of 0.5000 and a research size of $101.99. However, the decision was made not to use favorable bootstrap odds to override a failed holdout, absent a current signal or undetermined cash capture. A full fund exposure of $101.99 would be 12% of the account, and a quarter fund exposure would be 3% of the account, with the risk that stock loss can exceed the stop on a gap.

A model was run with 20,000 paths to analyze the performance of a trading strategy. The model used historical data from before 2024 and from the last 252 days. It considered fees of 1 basis point and a slip of 15 basis points per trade. The results showed that for the pre-2024 data, the stop loss was triggered first 66.77% of the time, the target was reached first 31.21% of the time, and the trade timed out 2.02% of the time. The mean net return was 1.50% and the median return was -7.01%. For the last 252 days, the stop loss was triggered first 50.78% of the time, the target was reached first 46.42% of the time, and the trade timed out 2.80% of the time. The mean net return was 5.78% and the median return was -7.01%. A half-Kelly strategy was also analyzed, which is a method of determining the optimal bet size, using a grid search from 0 to 1 and then taking half of the result. The strategy was tested on a training set of 49 trades, a holdout set of 10 trades, and a forward set of 7 trades. The results showed that the half-Kelly strategy had a stock weight of 0.3845 for the training set and 0.0000 for the holdout set. The decision was made not to use the favorable bootstrap odds to override the failed holdout, absent a current signal or undetermined cash capture. A full fund exposure of $101.99 would be 12% of the account size of $849.92, and a quarter fund exposure of $25.50 would be 3% of the account size.

A news scan found the report for R91806 TEN. The report is about an automated trading fund manager's notes. The model used 20,000 paths and 60 sessions to analyze historical data. It looked at open and close prices, as well as high and low prices, over 5-day blocks. The model also considered fees of 1 basis point and a 15 basis point slip per side. The results showed different outcomes for different time periods and strategies. For example, from before 2024, the model found that 72.725% of the time, the stop was hit first, and 26.645% of the time, the target was hit first. The model also calculated the mean net return and median return for each scenario. The report also discussed the concept of half-Kelly, which is a strategy for sizing trades, and found that the optimal stock weight was 0.0000 for the training set and 0.5000 for the holdout set. The report concluded with a decision not to use favorable odds to override a failed holdout, and noted that a full fund exposure would be 12% of the account, and a quarter fund exposure would be 3% of the account.

The source https://runway.adityaperswal.workers.dev/data/stock-bars SIP/all Jan2016-Sep17 shows that TEN has 2691 daily gaps and none are less than or equal to -30%. The worst gap was -9.5963% on April 29, 2020, and the worst pre-2024 close return was -17.58995%. This does not mean there is no risk of a future disaster. Using a simple model, the daily upper limit is 11.1262%, which is 6.46126% over 60 sessions. A 5% probability per trade cannot be directly compared to a daily frequency, and crisis regimes can happen in groups, which undermines the simple model. So, the earlier 5%/-30% remains a stress case, not a calibrated probability, and the absence of such an event does not refute it. PSIX had an actual adjusted gap of -34.18735% on May 12, 2026, and -29.99754% on November 7, 2025. LPG had a worst pre-2024 gap of -22.27513%. The AA feed on November 1, 2016, showed an apparent -74.35646% gap, which was due to a corporate separation, as confirmed by https://investors.alcoa.com/resources/investor-faqs/default.aspx. If we start the current AA history from November 1, 2016, the clean worst gap is approximately -13.2465%. The correlation between TEN and other stocks is: TEN/FRO is 0.819413, TEN/STNG is 0.708182, and TEN/LPG is 0.600194. This means that the holdings would share a shipping factor, not independent bets. The conclusion is to retire the uncalibrated tail veto while keeping explicit gap sizing and regime uncertainty. The sample of closed trades cannot estimate the edge. The hard account risk and fund 101.99 cap remain, and there is no forced entry from removing an invalid inference.

A diagnostic test was run on the stock ticker symbols TEN, PSIX, AA, CENX, ALM, and MU using a breakout strategy. The test used a computer program called node to run a script called backtest.js with specific settings, including a lookback period of 20 days and a stop loss of 8 percent. The data for the test came from the website https://runway.adityaperswal.workers.dev/data/stock-bars. The test did not take into account real-world costs like commissions or slippage, and it did not test how the strategy would work in real-time trading. The results for the stock ticker symbol MU showed a return of 33.27 percent, with a maximum drawdown of 36.24 percent and a win rate of 41.67 percent. The average trade return was 0.47 percent. These results are for diagnostic purposes only and do not validate the current eligibility of the strategy, as they are based on a different strategy and data feed.

A diagnostic test was run using the command line interface (CLI), which is a way to interact with a computer program using text commands. The test used a program called node to run a script called backtest.js with specific parameters: it looked at the stocks TEN, PSIX, AA, CENX, ALM, and MU, used a breakout strategy, looked back 20 days, tested over 1460 days, stopped losing trades at 8%, and aimed for a 24% gain. The data for the test came from https://runway.adityaperswal.workers.dev/data/stock-bars. The test assumed that trades were made at the same closing price, with no fees or losses from buying or selling, and did not account for any bad starts to trades. This test is only for diagnosis and not for making actual trades. Other analyses, which include costs and are split by time, are more important for making decisions. The results of this test are for a different strategy and use different data, so they cannot be used to confirm or deny the current plan. The full results for the stock ALM are: it was tested over 297 days, had a return of 124.51%, while buying and holding the stock would have given a 190.14% return, the biggest loss was 18.11%, 50% of trades were winning trades, and the average trade gave a 4.21% return.

A diagnostic test was run on the stock symbol CENX using a breakout strategy. The test used a command line interface, or CLI, with specific parameters: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback 20 --days 1460 --stop-pct 8 --target-pct 24. The data for the test came from https://runway.adityaperswal.workers.dev/data/stock-bars. The test assumed that trades were made at the same closing price, with fixed brackets, and without any commissions or slippage. It also did not account for any gaps in the market at the opening. The results of the test are for diagnostic purposes only and do not validate the current strategy. A different analysis using vectorbt, which includes costs, will be used for inference. The results of the test for CENX were: 1003 bars of data, a return of 30.78%, a buy and hold return of 433.07%, a maximum drawdown of 34.92%, a win rate of 32.5%, and an average trade return of 0.53%.

A diagnostic test was run on a trading strategy using the command: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback 20 --days 1460 --stop-pct 8 --target-pct 24. The data came from https://runway.adityaperswal.workers.dev/data/stock-bars. The test used a system called IEX, which looks at signals to enter trades when the market closes, uses fixed price ranges for trades, and does not account for costs like commissions or slippage. It also does not hold out any data for validation and assumes trades are executed at the same price as the signal, without any gaps in pricing at the opening. This test is only for diagnosis and not for validating the execution of trades. Other analyses, which include costs and are split by calendar, supersede these results for making inferences. The results are for a different strategy and data feed, so they cannot be used to confirm or deny the current eligibility. The full metrics for the symbol AA are: it had 1003 trading periods, a return of -25.39%, a buy and hold return of 5.35%, a maximum drawdown of 42.47%, a win rate of 32.86%, and an average trade return of -0.34%.

A diagnostic test was run on a trading strategy using the command line interface (CLI) with the following parameters: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback 20 --days 1460 --stop-pct 8 --target-pct 24. The data used for this test came from https://runway.adityaperswal.workers.dev/data/stock-bars. The test was run using the IEX engine, which uses same-close signals and entries, fixed brackets, and does not account for commissions or slippage. The test also did not include a holdout period or account for adverse opening-gap execution. The results of this test are for diagnostic purposes only and do not validate the current eligibility of the strategy. The results are also not comparable to other analyses, such as the vectorbt analyses 3310-3321, which supersede these results for inference. The full metrics for the symbol PSIX are: the symbol PSIX had 432 bars, a return percentage of 42.32466690454897, a buy and hold percentage of 51.500000000000014, a maximum drawdown percentage of 15.0166294688044, a win rate of 41.935483870967744, and an average trade percentage of 1.3974287736690292.

A diagnostic test was run on a trading strategy using the command line interface (CLI) with the following parameters: node /app/agent/dist/backtest.js --symbols TEN,PSIX,AA,CENX,ALM,MU --strategy breakout --lookback 20 --days 1460 --stop-pct 8 --target-pct 24. The data used for this test came from https://runway.adityaperswal.workers.dev/data/stock-bars. The test was run using the IEX engine, which uses same-close signals and entries, fixed brackets, and does not account for commissions or slippage. The test also did not include a holdout period or account for adverse opening-gap execution. The results of this test are for diagnostic purposes only and do not validate the current eligibility of the strategy. The results should not be compared to other analyses, such as the vectorbt analyses 3310-3321, which are more comprehensive and include costs. The full metrics for the symbol TEN are: it had 1001 bars of data, a return percentage of 23.779307240557834, a buy and hold percentage of 178.24367088607596, a maximum drawdown percentage of 19.727161495833574, a win rate of 37.333333333333336, and an average trade percentage of 0.39570689620069754.

The automated trading fund manager analyzed the daily stock market data from the source https://runway.adityaperswal.workers.dev/data/stock-bars. The manager used a statistical method called NumPy OLS to predict the stock returns over the next 20 sessions. This method used current breakout and moving average data, including SMA50 and SMA200, as well as 20-day momentum. The analysis was trained on data up to 2023 and did not include data from 2024 and 2025. The manager also used a technique called Bartlett HAC20 to address overlapping data. The results showed that for certain stocks, including TEN, FRO, STNG, LPG, AA, CENX, FTK, and MU, the breakout days were above the SMA50. However, these results were not independent and were strongly overlapping. The manager also analyzed the signal coefficients for these stocks, which measure the relationship between the stock prices and the predicted returns. The results showed that the signal coefficients for most of the stocks had confidence intervals that included zero, meaning that the results were not statistically significant. Additionally, six out of eight of the out-of-sample R2 values were negative, indicating that the model did not perform well on new data. The manager concluded that there was no statistically reliable extra independent signal from the confluence of these stocks and decided to retain the original hypotheses and collect more primary economic data.

This is a report on a trading strategy for the SPY stock. The strategy was tested using data from 2016 to 2023 and then applied to a holdout period from 2024 to 2025. A separate test was done for 2026. The strategy uses a breakout system based on the Average True Range (ATR), which is a measure of how much a stock's price moves, and the Simple Moving Average (SMA) over 50 days. The strategy also uses stop-loss orders and targets for selling. The results of the tests were: for the training period, a return of 110.5027% was marked and 108.0478% was closed, with a maximum drawdown of 15.8699% and a win rate of 50.00%. For the holdout period, a return of 25.6958% was marked and 24.8288% was closed, with a maximum drawdown of 6.4636% and a win rate of 45.45%. For 2026, a return of 0.5309% was marked and 0.3321% was closed, with a maximum drawdown of 8.4508% and a win rate of 20.00%. The stress test for the holdout period showed a return of 22.8380% marked and 22.1126% closed, with a maximum drawdown of 6.8370% and a win rate of 45.45%. The report notes that this is an experiment and not proof of a successful trading strategy.

This is a preregistered note, with detailed methods found in note 3577. The daily data was collected from September 17, with training from 2016 to 2023, and a calendar holdout from 2024 to 2025. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy used a next-open prior 20-close breakout and a 50-day simple moving average, which is a calculation to determine the average price of a stock over a certain period. The strategy also used an average true range, which is a measure of volatility, with brackets of 2 times the average true range below and 6 times above the average true range. The maximum holding period was 60 sessions. The strategy included entry-day stops, which are orders to sell a stock when it reaches a certain price, and adverse opening gaps, which occur when a stock opens at a lower price than the previous day's close. The target was to open at a priority price, and the stop-loss order was set to be ambiguous during the day. The simulation used a $10,000 investment, with a 1 basis point fee and a 5 basis point slippage per trade, and a stress test with a 15 basis point slippage. The results showed a return of 14.8198% during the training period, with a closed compound return of 15.8252%, a maximum drawdown of 69.0629%, and a win rate of 33.33%. The holdout period showed a return of 32.4452%, with a closed compound return of 28.6361%, a maximum drawdown of 56.5251%, and a win rate of 41.18%. The forward period in 2026 showed a return of 57.9976%, with a closed compound return of 54.7930%, a maximum drawdown of 36.9163%, and a win rate of 50.00%. The stress test during the holdout period showed a return of 27.8900%, with a closed compound return of 24.3360%, a maximum drawdown of 57.0437%, and a win rate of 41.18%. The results also included trend filters, such as the L10/40, which were recorded as sensitivities but not selected by the holdout. The conclusion is to retain this strategy as a research candidate, as historical positivity alone does not establish current signal, valuation, or executable payoff.

This is a preregistered note, detailed in methods3577. The daily data used is from September 17, with training from 2016 to 2023, a calendar holdout from 2024 to 2025, and a separate diagnostic for 2026. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy used is a next-open prior20-close breakout and a 50-day Simple Moving Average, with a 20-day arithmetic Average True Range, which is a measure of volatility. The strategy also includes entry-day stops and adverse opening gaps, with a target opening priority. The results are as follows: in the training period, the strategy had a marked return of -89.0755%, a closed compound return of -89.0755%, a maximum drawdown of 94.3601%, and a win rate of 16.67%. In the holdout period, the strategy had a marked return of -27.3629%, a closed compound return of -30.9049%, a maximum drawdown of 45.5460%, and a win rate of 26.32%. The forward 2026 results show a marked return of -29.2466%, a closed compound return of -29.2466%, a maximum drawdown of 32.3656%, and a win rate of 16.67%. The stress holdout results show a marked return of -30.1412%, a closed compound return of -33.4812%, a maximum drawdown of 46.7841%, and a win rate of 26.32%. The conclusion is that the baseline strategy lacks consistent support between the training and holdout periods, and is therefore rejected as a standalone live rule.

This is a report on a trading strategy for the stock CENX. The strategy was tested using daily data from September 2017, with training data from 2016 to 2023, a holdout period from 2024 to 2025, and a diagnostic period in 2026. The data was sourced from https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy uses a breakout system, where a trade is entered when the stock price breaks out of a certain range, and includes stop-loss orders and target prices. The strategy was tested using a $10,000 investment, with a 1 basis point fee and a 5 basis point slippage per trade. The results show that the strategy performed well in the training period, with a marked return of 106.4773% and a closed compound return of 118.1736%. However, the performance was lower in the holdout period, with a marked return of 55.3394% and a closed compound return of 29.6065%. The strategy also performed poorly in the 2026 diagnostic period, with a marked return of -28.4114%. The report concludes that the strategy should be retained as a research candidate, but notes that historical performance does not guarantee future results.

This is a preregistered note, detailed in methods3577. The daily data used is from September 17, with training from 2016 to 2023, and a holdout period from 2024 to 2025. A separate diagnostic test will be done in 2026. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The trading strategy used is a next-open prior20-close breakout and SMA50, with prior20 arithmetic Average True Range (ATR), which is a measure of volatility, and E-2ATR/E+6ATR brackets, which are price levels used to set stops and targets. The strategy also includes entry-day stops and adverse opening gaps, with a target opening priority and ambiguous intraday stop first. The Vectorbt backtesting framework was used with $10000 reinvested, a 1 basis point fee, and a 5 basis point slip per side. The results show that during the training period, the strategy had a marked return of 132.3927%, a closed compound return of 104.7072%, a maximum drawdown of 59.6778%, and a win rate of 35.14%. In the holdout period, the strategy had a marked return of 48.8489%, a closed compound return of 32.0076%, a maximum drawdown of 44.8167%, and a win rate of 36.36%. In the forward 2026 period, the strategy had a marked return of -4.7619%, a closed compound return of -4.7619%, a maximum drawdown of 21.9454%, and a win rate of 33.33%. The stress holdout results were similar to the holdout results. The L10/40 and other trend filters were recorded as sensitivities, but not selected by the holdout. The conclusion is to retain this strategy as a research candidate, but historical positivity alone does not establish current signal, valuation, or executable payoff.

This is a report on an automated trading fund manager's notes. The manager used a preregistered method, detailed in notes 3573 and 3577, to analyze daily stock prices from November 1, 2016, to September 17, 2023, and then held out the years 2024 and 2025 to test the strategy. The manager used a website, https://runway.adityaperswal.workers.dev/data/stock-bars, as the source of the stock prices. The strategy involved buying a stock when its price broke out above the previous 20-day closing price and the 50-day simple moving average, with a stop-loss order in place. The manager also used a measure called the average true range, or ATR, which is a way to measure the volatility of a stock's price, to set the stop-loss and take-profit levels. The manager simulated the strategy using a $10,000 investment and assumed a 1 basis point fee and a 5 basis point slippage per trade, which means the actual price paid for the stock was 0.05% worse than the expected price. The results of the simulation showed that the strategy performed well in the holdout period, with a marked return of 84.6740% and a closed compound return of 53.3232%. However, the strategy performed poorly in the forward period, with a marked return of -7.4838% and a closed compound return of -7.4838%. The manager noted that the results are not conclusive due to the short history and small sample size, and that the strategy is still unvalidated for allocation. The manager also noted that the results include open positions, but the closed statistics do not. A news scan found the headline was not available, this is a full report.

This is a report on the R91806 PSIX automated trading fund. The report is based on a preregistered note and detailed methods. The data used is from September 17, with training from 2016 to 2023, a calendar holdout from 2024 to 2025, and a separate diagnostic for 2026. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The trading strategy is based on a breakout and a 50-day Simple Moving Average, with a stop-loss and target price. The strategy was tested using a $10,000 investment with a 1 basis point fee and a 5 basis point slip per trade. The results for the holdout period are a marked return of 220.1909%, a closed compound return of 220.1909%, a maximum drawdown of 28.9480%, and a win rate of 100.00%. The results for the forward period in 2026 are a marked return of -10.2456%, a closed compound return of -15.5546%, a maximum drawdown of 42.6214%, and a win rate of 25.00%. The stress test results for the holdout period are a marked return of 217.6396%, a closed compound return of 217.6396%, a maximum drawdown of 28.9480%, and a win rate of 100.00%. The report notes that the results are based on a technical overlay experiment and not proof of alpha, and that the short history remains unvalidated for allocation.

This is a preregistered note, detailed in methods3577. The daily data used is from September 17, with training from 2016 to 2023, and a holdout period from 2024 to 2025. A separate diagnostic period is set for 2026. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy uses a next-open prior 20-close breakout and a 50-day simple moving average, which is a calculation of the average price of a stock over 50 days. The strategy also uses an arithmetic average true range, which measures the volatility of a stock, with brackets of 2 times the average true range below and 6 times above the average true range. The strategy has a 60-session maximum and includes entry-day stops and adverse opening gaps. The target opening priority is used, and ambiguous intraday stops are handled first. The Vectorbt backtesting framework is used with a $10,000 reinvested fractional proxy, a 1 basis point fee, and a 5 basis point slip per side. A stress test with a 15 basis point slip is also performed. The fixed brackets do not include 2R management, which refers to a risk management strategy, and actual 15-minute and overnight execution. There is no pre-2024 training history. The holdout period showed a marked return of -16.8897%, a closed compound return of -20.6682%, a maximum drawdown of 32.2133%, a win rate of 14.29%, 7 trades, 1 open trade, and 1 target. The forward period in 2026 showed a marked return of 122.8193%, a closed compound return of 112.7180%, a maximum drawdown of 10.9551%, a win rate of 83.33%, 6 trades, 1 open trade, and 5 targets. The stress holdout period showed a marked return of -18.1270%, a closed compound return of -21.7711%, a maximum drawdown of 32.9548%, a win rate of 14.29%, 7 trades, 1 open trade, and 1 target. Other trend filters, such as L10/40, were recorded as sensitivities but not selected by the holdout. The current survivor selection, retrospective adjustments, small sample size, prior viewing of history, and no historical primary-event labeling mean this is a technical overlay experiment, not proof of bottleneck alpha. The marked return includes open paper, while the closed statistics do not. The conclusion is that the short history remains unvalidated for allocation.

This is a preregistered note, detailed in methods3577. The daily data was collected from September 17, using 2016-2023 as the training period, which started on November 1, 2016, 2024-25 as the calendar holdout, and 2026 as a separate diagnostic period. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy used a next-open prior 20-close breakout and a 50-day Simple Moving Average, with a 20-day arithmetic Average True Range, which is a measure of volatility, and brackets of 2 times the Average True Range below and 6 times above the Average True Range. The strategy also included entry-day stops, adverse opening gaps, and a target opening priority. The Vectorbt backtesting framework was used with a $10,000 reinvested fractional proxy, a 1 basis point fee, and a 5 basis point slip per side, which is a small cost added to each trade. The results for the training period were a marked return of 19.9143%, a closed compound return of 27.7017%, a maximum drawdown of 78.3639%, a win rate of 30.61%, 49 trades, 1 open trade, and 13 targets. For the holdout period, the results were a marked return of -25.6595%, a closed compound return of -25.6595%, a maximum drawdown of 40.4370%, a win rate of 20.00%, 10 trades, 0 open trades, and 2 targets. The forward 2026 results were a marked return of 90.8954%, a closed compound return of 76.5638%, a maximum drawdown of 11.9943%, a win rate of 57.14%, 7 trades, 1 open trade, and 4 targets. The stress holdout results were a marked return of -27.1315%, a closed compound return of -27.1315%, a maximum drawdown of 41.0297%, a win rate of 20.00%, 10 trades, 0 open trades, and 2 targets. The L10/40 and other trend filters were recorded as sensitivities but not selected by the holdout. The conclusion is that the baseline lacks consistent train and holdout support, so it is rejected as a standalone live rule.

This is a preregistered note, which means it was planned in advance, with detailed methods explained in note 3577. The trading strategy was tested on daily stock prices from September 2017, using data from 2016 to 2023 for training, 2024-2025 as a holdout period, and 2026 for diagnostic testing. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy used a breakout method, which means buying when the stock price goes above a certain level, and also considered the average true range, or ATR, which is a measure of price volatility. The strategy also used moving averages, specifically the SMA50, which is the average price over the last 50 days. The strategy had rules for when to enter and exit trades, including stop-loss orders, which are used to limit losses. The results were tested using a simulated $10,000 investment, with a small fee of 1 basis point, or 0.01%, and a slip, or difference between the expected and actual trade price, of 5 basis points. The results showed a loss of 34.85% in the training period, and a gain of 7.03% in the holdout period. The strategy was also tested with a stress test, which simulates difficult market conditions, and the results showed a gain of 5.12% in the holdout period. The results also showed that the strategy had a drawdown, or maximum loss, of 77.56% in the training period, and 26.65% in the holdout period. The win rate, or percentage of successful trades, was 28.57% in the training period, and 33.33% in the holdout period. The number of trades, or N, was 42 in the training period, and 9 in the holdout period. The strategy was not selected based on trend filters, and the results are considered a technical overlay experiment, rather than proof of a successful trading strategy. The conclusion is that the baseline strategy lacks consistent support and should not be used as a standalone live trading rule.

This is a preregistered note, detailed in methods3577, using daily data from September 17, with training from 2016 to 2023, and a holdout period from 2024 to 2025. The data source is https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy uses a breakout system, which means buying when the price goes above a certain level, and a stop-loss, which means selling when the price falls below a certain level. The strategy also uses a moving average, which is a way to smooth out price fluctuations, and an Average True Range, or ATR, which is a measure of price volatility. The strategy was tested with $10,000 and a small fee of 1 basis point, or 0.01%, per trade, and a slip of 5 basis points, or 0.05%, per trade. The results were: a loss of 77.28% in the training period, a loss of 10.76% in the holdout period, and a gain of 85.92% in the forward period. The strategy also had a maximum drawdown, or loss, of 84.59% in the training period, and 41.51% in the holdout period. The win rate, or percentage of winning trades, was 21.57% in the training period, and 36.36% in the holdout period. The strategy was not selected based on trend filters, and the results are not conclusive due to the small sample size and other limitations. The conclusion is that the baseline strategy does not have consistent support and should not be used as a standalone live rule.

The automated trading fund manager analyzed a costed ATR breakout strategy with a calendar holdout. The manager preregistered a note with detailed methods. The analysis used daily data from September 2017, with a training period from 2016 to 2023, a holdout period from 2024 to 2025, and a separate diagnostic period in 2026. The data source was https://runway.adityaperswal.workers.dev/data/stock-bars. The strategy used a next-open prior 20-close breakout and SMA50, with an arithmetic ATR, and E-2ATR/E+6ATR brackets, with a 60-session maximum. The manager included entry-day stops and adverse opening gaps, with a target opening priority and an ambiguous intraday stop first. The analysis used a Vectorbt $10,000 reinvested fractional proxy, with a 1bp fee and 5bp slip per side, and a stress test with a 15bp slip. The results showed a train period with a marked return of -68.7795%, a closed compound return of -68.9770%, a drawdown of 86.1382%, and a win rate of 19.57%. The holdout period showed a marked return of 13.3817%, a closed compound return of 13.3817%, a drawdown of 29.8290%, and a win rate of 37.50%. The forward 2026 period showed a marked return of 96.7260%, a closed compound return of 85.0282%, a drawdown of 22.0055%, and a win rate of 80.00%. The stress holdout period showed a marked return of 11.5820%, a closed compound return of 11.5820%, a drawdown of 30.2487%, and a win rate of 37.50%. The manager recorded L10/40 and other trend filters as sensitivities, but did not select them by holdout. The conclusion was that the baseline strategy lacked consistent train and holdout support, and was rejected as a standalone live rule.

The fund manager made some premises about PSIX, which is a company. First, they looked at the company's cash flow, which is the money coming in and out of the company. They found that the company's operating cash flow was 75.724 million dollars, and after taking out capital expenditures, which are big purchases the company made, they had 72.988 million dollars left. Then, they subtracted the 22.687 million dollars the company owed to customers and were left with 50.301 million dollars. The company's latest report does not include any gain from refunds. A refund is when a company gives money back to customers. The fund manager also looked at the company's stock price and found that on September 17, the price was 45.55 dollars per share, which means the company's total value was around 1049.9275 million dollars. The company's debt, which is the money it owes, was 72.6 million dollars, but it had 70.062 million dollars in cash, so its total value, including debt, was around 1052.4655 million dollars. The fund manager also found that the company's cash yield, which is the return on investment, was 6.9517 percent, but this is not an annual yield, which means it's not for the whole year. If they annualize the company's net income, which is its profit, they get a price-to-earnings ratio of 21.7277, but this is just a sensitivity test and not a forecast. A sensitivity test is a way to see how something changes when you change one of its parts. The fund manager also looked at the company's suppliers and customers and found that one supplier used to be very important to the company, but now it's not as important. The company still has one big customer, though. The fund manager concluded that the company's cash is better, but there are still risks, such as customer claims, that prevent them from considering the cash as a permanent rent. A counterargument is that if the company can normalize its backlog, which is the list of orders it has to fill, it could restore its margins, which are the profits it makes on each sale. The fund manager also looked at ALM, which is another company, and found that it has a new contract that improves its route-to-market, which is the way it sells its products. However, there are still some unknowns, such as the floor price, which is the minimum price the company will sell its products for, and the annual tonnes, which is the amount of product the company will sell. The fund manager cannot calculate the incremental NAV, which is the increase in the company's value, because of these unknowns. A news scan found the headline about the contract, but the fund manager did not verify the claims made in the headline. The fund manager will not make a speculative target from the contract headline. The binding constraints, which are the limits on what the company can do, are profit conversion and defensible valuation, which means the company needs to make a profit and have a reasonable value. The facts only go up to the disclosed periods, which means the fund manager only has information up to a certain point in time.

The report discusses how certain companies capture bottleneck rents. Firstly, a document from ALM on September 17 describes an agreement with WBH and Sandvik, which includes a minimum of 1720 tons of tungsten trioxide, a conditional 3 million USD upfront payment, and an index price or floor. This agreement allows for customer-funded supply and constrained producers. Secondly, a report from PSIX on August 6 states that power orders are facing production constraints, and the company's concentration of suppliers and customers may affect its ability to capture rents. Thirdly, an update from EGA on August 26 mentions that the restoration of its Al Taweelah smelter has reached a 25% completion milestone, with 315 out of 1262 cells restarted, and full output is forecast for Q1 2027. Additionally, a labor threat at MU is not confirmed to result in output loss, and the next capacity forecast is expected in 2027. The report notes that higher prices may not always benefit producers if their own volume falls or contracts are fixed. The facts presented are based on dated operations and contracts, and return causality and future margins remain assumptions.