The lab
We take a popular idea, write the rule down before computing anything, test it on years of minute data with real costs and publish the result — even when it comes out at zero.
How we test
- The rule is written down before the result is computed (no fitting after the fact).
- Minute data since 2012; one period to build, a separate one to check.
- Cost = spread and commission measured on real trades.
- A comparison with random entries — to tell skill from luck and from the market trend.
- Works Do US indices still tend to rise on the day before a market holiday?
In our data, yes: on US500, US100 and US30 with a broker, +40 to +66 bp on average for the session before a holiday after costs (t 3.4–4.0), and +44 bp on the S&P 500 in 1990–2010 (t 3.27).
- Fails after costs Do exchange data (who buys aggressively, order book imbalance, options) predict gold over the next 15–60 minutes?
Not in our tests. The opportunity exists (moves after such signals are about 3 times the cost on micro contracts), but none of the three measures pointed the direction: options flow was right 49.8% of the time.
- Works After the most important US releases, does gold continue the move of the first 30 minutes?
After FOMC, CPI and NFP — in our data, yes: +5.7 bp per trade after costs (t 2.66, 480 events). After 9 other releases (PCE, PPI, retail sales, GDP, ISM, JOLTS, ADP, durable goods) the effect disappeared after 2019.
- Fails after costs Does a filter of “only with the current session’s direction” improve short-term gold trades?
No. The filter cut about two thirds of the trades, but the rest did the same or worse: −0.91 and −1.21 bp per trade versus −1.02 bp without the filter (2022–2024).
- Fails after costs Do simple price rules (breakouts, rejections, VWAP) give an edge when scalping gold?
No. Of 10,680 rules, none passed the test in 2012–2019 or in gold’s current regime in 2021–2024. The median result before costs was +0.05 and +0.01 bp per trade — zero.
- Fails after costs Do popular MT5 indicators give an edge in short-term gold trading?
Not after costs. Of 1,144 combinations (EMA, MACD, RSI, Stochastic, CCI, Ichimoku, Bollinger, SAR, ADX…) none had a positive net result. The best one before costs made +0.40 bp per trade, and −1.73 bp on new data.
Past results do not guarantee future results. This is research, not a recommendation to trade.