Measured, Not Assumed — Tax-Aware Options Agent

Most premium-selling agents share two assumptions. We measured both, and both were wrong. First: everyone sells when implied volatility is high. Bucketing 2,406 days from 10.6 years of CBOE VIX data by IV rank at entry, the mean premium earned was +4.27 vol points below rank 30 versus +0.99 above rank 70 — four times better at LOW IV rank. The median runs slightly the other way, so high IV rank is not selecting losing trades; it is selecting trades with much worse tails. We keep IV rank as a risk floor, not an alpha signal. Second, and this changed the strategy: retail backtests model option slippage at $0.02/share. We measured 2,170 real SPY option quotes on the live chain — the median half-spread is $0.17, eight times higher. Re-running the same 8.6-year backtest, changing only that constant, with identical trades and identical gross P&L of +$3,791, a $5-wide spread strategy goes from +$2,741 to -$3,199. The sign flips. The toll is fixed per contract while credit scales with width, so narrow spreads pay the same fee to collect less. The agent now trades wide and few, and refuses any trade that cannot clear $150 net of the execution cost. Two things we have not seen elsewhere. An IRC 1091 wash-sale guard blocks re-entry for 30 days after a realised loss, so the loss stays deductible. And the books reconcile to Alpaca every cycle: quantity-aware position matching, credits trued to actual multi-leg fills, positions marked at liquidation value. Across 18 recorded cycles, the drift between our equity and the broker's averaged $4.42 on a $100,000 account. verify_books.py exits non-zero if they ever disagree. Two days of paper trading is not evidence that a strategy works, and we do not present it as such. What we stand behind is that every constant is measured rather than assumed.