Van Tharp and Tom Basso ran a famous test: enter long or short at random — literally a coin flip — across 10 markets. It still made money every time, once they added just two things: a trailing stop (an exit that follows the price to lock in gains and cap losses) and sensible position sizing (risking only about 1% of the account on each trade). The win rate was only ~38%, yet it was reliably profitable. The lesson is blunt: the exits and the bet-size carried the edge, not the entry.

We ran 106 backtests on COIN (Coinbase stock — a high-beta crypto proxy that swings about 3.5× the market). We found a trend strategy (a Donchian channel — buy on a break above the recent high, sell below the recent low) that beats simply buying and holding, on paper: a profit factor of ~1.39 (win $1.39 for every $1 lost) with a max drawdown of only ~6%.

Then came the honest lesson: a "profit factor above 2 on a single symbol" is essentially a mirage — a number you can only reach by overfitting (tuning so tightly to the past that it fails the moment it goes live). The literature is blunt: real standalone edges are modest, and anything spectacular is either a secret or an artifact. So we re-pointed the lab at what the math actually rewards: a diversified basket plus volatility-targeted sizing.

To be clear: this is paper / research only. Nothing here is live, and none of it is a recommendation to buy, sell, or hold anything. Promoting any strategy to real money is a deliberate human decision — never automatic.

Van Tharp 同 Tom Basso 做過一個著名測試：喺 10 個市場隨機做多或做空 — 真係擲公字咁決定 — 結果次次都賺錢，前提只係加咗兩樣嘢：移動止損（跟住價格走嘅出場，鎖住利潤、封住蝕損）同合理嘅注碼管理（每筆只冒帳戶約 1% 風險）。勝率得約 38%，但照樣穩定有錢賺。教訓好白：帶嚟優勢嘅係出場同注碼，唔係入場。

我哋喺 COIN（Coinbase 股票 — 高貝塔加密替代品，波幅約大市 3.5 倍）跑咗 106 次回測。搵到一個趨勢策略（唐奇安通道 — 突破近期高位就買、跌破近期低位就沽）喺紙上贏過純粹買入持有：獲利因子約 1.39（每蝕 $1 賺 $1.39），最大回撤只係約 6%。

跟住就係老實嘅一課：「單一股票獲利因子高過 2」基本上係海市蜃樓 — 一個只能靠過度擬合（調到太貼歷史，實戰即失效）先達到嘅數字。文獻講得好白：真實嘅單一優勢好溫和，任何誇張數字唔係機密就係假象。所以我哋將實驗室重新指向數學真正獎勵嘅方向：分散化籃子加波動率目標注碼。

講清楚：呢啲全部係紙上／研究性質。冇一樣係實盤，亦冇一樣係叫你買、賣或持有任何嘢嘅建議。任何策略上實盤，都必須由人手刻意決定 — 永遠唔會自動。