Massive

Many thanks for the kind words @JNandezPoker 🤍 I’ll answer as one description to avoid overlap. As part of the original backtesting and modelling work, I used data from @massive_com and walked forward using only the market cap and D$V with the eligible tickers dynamically selected so that delisting, acquisitions, mergers etc were included. The work was then repeated using the current universe for comparison, which does carry survivorship bias. However, while not negligible, the trades related to these stocks did not materially change the model. For example, I would need to be either: i) in the trade when the acquisition, merger, delisting occurred and or close enough to the event that it impacted the stock price or ii) the trade was completed outside of those events and fell within the normal range of all other trades. The chart line is a simulation with capacity constrained, not live fills. It is the current rule set run back over the whole period using @massive_com data, walked forward day by day with the eligible universe filtered on market cap ($1B+) and dollar volume, so each day only sees what was known that day. It is capacity constrained with no leverage: 2% per position, max 50 positions. Every eligible trade (buys, adds, trims, sells) is captured, then a Monte Carlo randomizes which trades get taken when capital is full, over hundreds of runs, to give the P10 and P90 trader (the unlucky and lucky versions). It has also been tested with a ranking layer deciding which trades to take when capital is constrained. On in-sample vs out-of-sample: the current rules were mostly finalized before the start of 2026 however any areas of improvement and new rules are always incorporated after verification and modelling with new backtests. It was always intended that the "model" is the best version of the algorithm at that point in time in conjunction with perfect execution of by the trader. This will always be the most important "slippage" and the goal is to strive towards full alignment and automation to minimize that slippage. Since January 2026 the signals have been traded live with part of the capital, selected manually and sized at 1% per position, executed into the close to match the model. Full API automation is in beta until year end, with confirmation required before each order. With respects to costs, using modern low-cost brokers the fees are zero on common stocks. Speaking in real terms the fees on the model trades, they were 0.002%. For slippage, I find this to be an over-blown consideration when you are trading only liquid names with $1B+ market cap stocks and D$V of $50M+. The orders are placed as limit or "buy on close" orders to the broker via API and automatically update/feedback/revised every few seconds based on the last trade price. Some orders are filled a few cents above the close and some are a few cents below the close. They tend to net out on balance. Finally, there are two presentations of the algorithm at present: 1) The simulation/model that walked forward capacity constraints from a particular start point simulating perfect execution. The positions that model holds can depend on the current market conditions in a sense that if there were too many opportunities it may have "skipped" some trades as it was allocated fully when the signals occurred. 2) The all open-trades without capacity constraints showing the currently eligible buys and the holdings. During bear markets, the positions of the two views will align fully but during bull markets, the "all open-trades' view will have more positions than the model/simulation. There is also this view of the same model (pic 1). I asked Claude to reconcile the two views (see pic 2). The TradingView version is a new development as I thought it might prove instructive.