A while ago, I wrote about 我如何为FOMC周构建AI宏观仪表盘 —— 我在其中写到了我如何使用AI将经济日历、新闻和情景压缩成真正可以用来交易的信息。那篇文章讲的是保持信息了解和做好准备。这篇文章讲的是之后发生的事情:一个令人不安的现实——即使您有不错的宏观判断和可靠的仪表盘,仍然可能在执行环节漏钱。
START
READ "trade_journal.csv"
COLUMNS: date, symbol, pnl, notes
DEFINE EMOTION_SETS:
impatience = ["should move", "too early", "soon", "jumped in"]
overconfidence = ["high conviction", "obvious", "easy trade"]
fear = ["scared", "nervous", "hesitated", "cut early"]
revenge = ["get it back", "make it back", "chase", "missed it"]
hope = ["it will come back", "just this once", "give it room", "macro is still good"]
FOR EACH trade IN journal
text = LOWERCASE(trade.notes)
themes = EMPTY_SET
FOR EACH (label, phrases) IN EMOTION_SETS
FOR EACH phrase IN phrases
IF phrase IS IN text THEN
ADD label TO themes
ENDIF
ENDFOR
ENDFOR
IF text CONTAINS ANY OF ["early", "no trigger", "moved stop", "oversized", "cut early"] THEN
error_type = "execution"
ELSE IF text CONTAINS ANY OF ["wrong thesis", "missed macro", "data changed", "read was wrong"] THEN
error_type = "analysis"
ELSE
error_type = "unclear"
ENDIF
ATTACH themes AND error_type TO trade
ENDFOR
COUNT how many times each theme appears across all trades
OUTPUT "emotion_summary.csv" WITH COLUMNS: theme, count
OUTPUT "journal_flags.csv"
COLUMNS: date, symbol, pnl, themes, error_type, notes
END
就这么简单,没有魔法。
这里的“AI”只是我用来帮助总结和标记这些主题的 Claude code 模型——真正的工作仍然在于我所写的内容,以及我决定执行哪些规则。
请注意,我并非想提出一个全新的理念!“时机、仓位、心理”是交易中最古老的三要素。
我写这篇文章的原因更简单:我需要一种方法,让这三个问题在我自己的盈亏 (P&L) 中无法被忽视。宏观仪表盘能帮我在大局上做到“大致正确”。而这种日志加 AI 的工作流程,则是我用来发现自己在实践中仍然犯下的所有错误,并将这些漏洞转化为我真正能够遵循的规则的方法。