解锁这篇笔记

A deep discount in Taiwan index futures is usually arithmetic, not fear. Every June through September, Taiwan's weighted index sheds hundreds of points to constituent dividends — about 592 points in 2026, with a single TSMC ex-dividend date worth roughly 50. Futures price that erosion in ahead of time, so the "deep backwardation" on your screen is mostly dividends that haven't dropped out of the index yet. This strategy (TXF breakout + mispricing-convergence dual signal, official and free) strips the dividend points out first, then decides whether the future is genuinely cheap — and pairs that with a trend-breakout leg. Backtested 2018–2026: +177.6% total return, 12.6% annualized, Sharpe 1.20, max drawdown −19.6%. Two independent long signals The strategy runs on 60-minute bars; the two legs enter and exit on their own, and positions add. Breakout leg: go long when the close breaks the 250-bar high, with an ATR×2 trailing stop that only ratchets up — the trend decides how long you stay. Mispricing leg: each day, restore "futures minus spot" to true mispricing — subtract cost of carry, add back the dividend points that will still evaporate between today and settlement (synthesized from announced ex-dividends plus historical realizations). Take a 252-day z-score of that clean residual: enter below −1.5 to ride the convergence, exit at −0.5. Why can't you read the raw basis directly? Because the raw spread mixes three things: carry, pending dividends, and actual mispricing. In dividend season, a discount of tens to hundreds of points is simply what fair value looks like — treat it as a panic gauge and it misleads you systematically, at the same time every year. This strategy only acts on the residual after everything else is stripped out: cheap for real, not cheap-looking. An honest detour: fixing the spread made the backtest worse We assumed correcting for dividends would improve performance. It didn't — the corrected signal earned less, because a large chunk of the raw-spread profit turned out to be seasonal timing around dividend season, not convergence itself. We kept the correct spread definition anyway (signal quality first) and recovered the performance from the breakout leg's exit — replacing a fixed channel exit with the ATR trailing stop. Backtest performance (2018–2026, 60-minute bars) MetricValueTotal return+177.6%Annualized12.6%Sharpe1.20Max drawdown−19.6%Trades692MCPT p-value0.0025 (only 5 of 2,000 random shuffles did better) Fees of 0.01% per side included; flat about 80% of the time. Backtest results, not a guarantee of future performance. Three things to know before you run it First, the mispricing leg is not strongly significant on its own — its value rests on economic logic and its diversification contribution. Second, dividend points more than two weeks from settlement partly rely on estimates (announced coverage is about 44% at 17 days out, 100% at 4 days), so early-season entry timing can differ slightly from the backtest. Third, each leg sizes with a volatility target (2× cap per leg); with both legs in the market, instantaneous combined exposure can reach 4× — know what you're carrying before you deploy. The strategy is free. The hard part is keeping it running. The logic is above. But synthesizing dividend points daily, computing z-scores, watching 60-minute breakouts and trailing stops, and routing orders without missing a bar — that's the infrastructure traditional quant burns weeks on. Blave Agent runs that chain for you. You're not paying for the strategy; you're paying to keep it live. Start in 3 steps ① Install Blave Agent② Add credit (from 300 TWD) to start your Blave Agent③ Tell it "load the official TXF dual-signal strategy" → one-click deploy, runs on schedule You don't build it yourself — the official strategy is free, one line loads it. From 300 TWD to start; then you pay only for runtime: server from 1.6 TWD/hr, platform 3 TWD/hr, plus AI usage. Install Blave Agent and put this strategy to work today →