Should a swing position be closed before the weekend? Testing Friday exits, hunting an intraday edge in stocks, and a silver evening lead

Only on medium gaps. The stand-down was written for openings of 0.2–0.5%, where a stretched index and a modest gap together averaged a loss in the backtest. Openings of half a percent or more are a different population — the gap itself is the signal, and the backtest kept those regardless of stretch. So the bot did what its rules say. Reading a rule's scope from its exact trigger, not from its name, is most of what a bot operator does.

The weekend carry was worth about a fifth of the system's total over 2020–2026, and overnight carry in general about three-fifths — the system is, in large part, a holder of gaps. Every Friday rule we pre-registered failed: close on Friday, don't enter on Friday, don't enter from Wednesday on, close only when the position is below a threshold — each lost between 55 and 73 percent of the profit, because the trades that survive to a Friday are the trending ones, and trends do not respect the calendar. The honest number on the other side: about 27 percent of weekend carries gapped through the stop on Monday morning, so the weekend is where the biggest single losses live. The decision is to keep carrying, with the cost stated.

The engine says no, and so does the all-time record of single lots: the best trailing lots in the backtest ran for weeks and made several times the typical trade, and they are what the whole edge is made of. The floor is what matters: with the stop sitting at the last pivot low, a gap down of more than about two and a half percent would be needed to lose more than roughly half of the open gain, and that size of overnight gap has happened on a few percent of trading days. The rule that manages this — the stop steps up only to a new pivot low, never down — has already been tested against every alternative. Changing it by hand because the number on the screen is large is exactly the discretionary move the system exists to prevent.

It was run both ways — long and short, intraday, on 5-minute bars, with realistic cash-market costs — on all fifty. It fails on every one and on the basket: the short side loses in most years, the long side is a coin toss, and the few stocks that look positive do not survive splitting the sample in half. The setup is an index setup; the index has a smoother intraday structure than any single stock, and the leverage a buyer gets from a cheap option on the index is what makes the small average move pay. On cash stocks the same move is too small to cover the costs.

All of them were run with identical code and pre-registered bars: NIFTY, BankNifty, gold, gold mini, silver, silver mini, and the stocks, both sides. The long side on the index is the known result. The one new thing: on silver — big and mini — the long version of the setup earns consistently in the evening session, when the US market is open, in seven of eight years on silver mini and eight of nine on silver, with tiny drawdowns relative to the profit. Gold fails. Every short side on metals fails. The evening split was a post-hoc observation, so it is not a finding yet; it is a lead, which is why it went to a paper record rather than to money.

From five in the evening it watches 5-minute bars of the active silver-mini future. When a bar closes above a green supertrend with RSI above 70, above the previous day's first resistance pivot and above the upper Bollinger band, it records a one-lot buy at the quoted ask. It exits on the supertrend turning red or at the session's end, whichever is first, at the quoted bid, at most three entries an evening, with exchange charges applied. It also quotes the nearest in-the-money call at entry and exit, because the eventual question is whether the futures edge survives as an option. It places nothing; it is a measuring instrument for fills and for whether the backtest's evening pattern repeats forward.

The exchange margin for one lot is about sixteen percent of the contract's notional value — for a five-kilogram mini contract near ₹9.5 lakh, around ₹1.6 lakh. Intraday and overnight margin requirements turned out to be the same, and intraday positions are force-closed by the broker before the session's end, well before this system's own 23:15 exit — so a real version would have to be carried as an overnight-type position even though it never holds overnight. An option on the same contract would need only its premium, which is the whole reason the paper bot is recording option quotes.

No, in any form: base rules, advanced rules, full session or evening, the short side on silver is flat to negative, as it is on gold. Metals in this period trended up with sharp, short dips; a breakdown setup keeps buying dips that reverse. The asymmetry is the same one the index shows, only stronger.

Because the system's profit is concentrated in regimes, not years, and regimes do not repeat on a calendar. Measured by realised volatility and trend strength, this year sits in the top third of the sample for both, and the years that look like it are the ones that paid best. Matching tells you which past years are relevant comparisons — useful for setting expectations — but a plan built on them assumes the regime persists, which is the one thing a regime never promises. The baseline that is defensible is the full sample with its bad years in it, and a stated conditional: in a high-volatility, trending regime expect the upper half of the distribution; in a flat one expect the lower.