🕸️
MQL5 中的范畴论 (第 15 部分)：函子与图论·综合运用
（3/3）·把日历新闻连成图、用函子跨范畴配对标普波动，这套抽象框架到底能给交易带来什么
把财经日历只当事件列表刷，等于浪费了它内部的依赖结构。用图论把新闻项连成带环的网，再借函子投到波动率范畴，才可能看出线性排序抓不到的跨项传导。
◍ 负向投影的正向校准与相关性落差
在 SPX500 月线（MN1）上跑 morphs 预测，2023.05 给出的投影是 763492.32，而实际值是 28877.00；一个月后投影变成 -30298.31，实际 23306.00。两者相关性只有 0.034，morph 常数被设为 50000.00，说明这套形变模型在该品种月线尺度上几乎失去线性解释力。 为了不让负投影直接污染后续计算，代码里用 NormalizeForecast 把负值折回正域。逻辑是：以当前实际值为锚，按 (当前实际 - 负投影)/当前实际 的比例，去缩放上一期实际与上一期投影的绝对差，再加回当前实际。 换用 objects 预测后，相关性升到 0.46195608（morph 常数 0.00）。2023.05 投影 19185.80、实际 28877.00，2023.06 投影 11544.54、实际 23306.00——方向大致对，但绝对误差仍偏大。外汇与贵金属品种若套同逻辑，因杠杆与跳空风险高，月线投影失准可能放大仓位误判，建议先在 MT5 用历史数据复算 corr 再决定是否采信。
MQL5 / C++
class=class="str">"cmt">//+------------------------------------------------------------------+ class=class="str">"cmt">//| Normalizes negative projections to positive number. | class=class="str">"cmt">//+------------------------------------------------------------------+ class="type">class="kw">double NormalizeForecast(class="type">class="kw">double NegativeValue,class="type">class="kw">double CurrentActual,class="type">class="kw">double PreviousActual,class="type">class="kw">double PreviousForecast) { class="kw">return(CurrentActual+(((CurrentActual-NegativeValue)/CurrentActual)*fabs(PreviousActual-PreviousForecast))); }
形态映射在标普月线样本上的偏差
把一段形态映射（Morphs）脚本挂在 SPX500 的月线周期上跑，日志直接暴露了预测与现实的裂口。2023.05.01 零点形态预测值约 257937.93，而当时实际报价仅 28877.00；次月预测值冲到 679078.56，实际却落在 23306.00。 更关键的是相关性输出：correlation 为 -0.03589660，morph constant 是 0.00000000。负向且接近零的相关，加上常数为零，说明这套映射在该样本里基本没有解释力。 外汇与贵金属市场波动结构不同，但这类零相关结果提醒我们：任何形态外推模型在换周期、换品种后都可能失效，开 MT5 把同一脚本挂到 XAUUSD 月线对比一下，比盲信预测值更实在。
MQL5 / C++
class="num">2023.07.class="num">30 class="num">14:class="num">45:class="num">57.449 ct_15_w_n(SPX500,MN1) class="type">void OnStart() Morphs forecast is: class="num">257937.92970615, on: class="num">2023.05.class="num">01 class="num">00:class="num">00, with actual being: class="num">28877.00000000 class="num">2023.07.class="num">30 class="num">14:class="num">45:class="num">57.527 ct_15_w_n(SPX500,MN1) class="type">void OnStart() Morphs forecast is: class="num">679078.55629755, on: class="num">2023.06.class="num">01 class="num">00:class="num">00, with actual being: class="num">23306.00000000 class="num">2023.07.class="num">30 class="num">14:class="num">45:class="num">57.527 ct_15_w_n(SPX500,MN1) class="type">void OnStart() correlation is: -class="num">0.03589660 with morph constant as: class="num">0.00000000
「函子系数怎么挪到别的货币对去用」
范畴论在金融里最被低估的一点，是它能把一次 heavy 优化得到的映射系数直接搬到别的标的。上面测试里算出的对象函子映射和态射映射系数，若覆盖 MT5 策略测试器很长历史区间，计算开销极大；但系数一旦出来，套到欧元财经日历对 EURO-STOXX 波动率、或 FTSE 波动率对英镑的影响上，按假设可能比平均结果更好。 我们用德国范畴配 DAX30 波动率范畴跑了一遍，GER30 月线日志里 2022.09 预测 144734 实际 154697、2023.01 预测 440957 实际才 55544，末尾 correlation 只有 0.0226；UK100 那边 2023.01 预测 153576 实际 30453，correlation 0.1441。两个都低得离谱，说明策略测试器自带的日历数据太受限，假设根本立不住。 别把 0.46 当圣经 美元和 SP500 早先那 0.46 相关性只是极早期指标，真实可靠性取决于域与协域的线性系数。长远看，把财经日历手动导成 CSV 或塞进数据库，让智能系统读，才能在理想长区间里正经测。外汇和贵金属波动剧烈，这种迁移假设失效时亏损可能很快放大。 除了预测波动率定止损，持仓规模也能用同一套：把每笔交易结果当对象、结果随时间的演变当态射，通过函子映射到「理想仓位」范畴。亏了就按当时波动率等比缩仓，赚了等比放大，映射天然带滞后——这周期结果连下周期仓位。 映射别只写线性方程 基础做法是域值加常数乘系数得协域，但二次关系只需多一个平方项系数。更稳的是用人工神经网络同时学对象和态射的同态映射，多输入单输出天生合适，大量训练后可能比手拍系数准得多。
MQL5 / C++
class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">25.655 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">75550.00000000, on: class="num">2020.01.class="num">01 class="num">00:class="num">00, with actual being: class="num">210640.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">25.727 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">210640.00000000, on: class="num">2020.02.class="num">01 class="num">00:class="num">00, with actual being: class="num">431320.00000000 ... class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">28.445 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">144734.23693128, on: class="num">2022.09.class="num">01 class="num">00:class="num">00, with actual being: class="num">154697.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">28.539 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">171908.69769099, on: class="num">2022.10.class="num">01 class="num">00:class="num">00, with actual being: class="num">156643.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">28.648 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">298789.42486333, on: class="num">2022.11.class="num">01 class="num">00:class="num">00, with actual being: class="num">99044.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">28.753 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">178962.89317906, on: class="num">2022.12.class="num">01 class="num">00:class="num">00, with actual being: class="num">137604.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">28.901 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">440957.33259197, on: class="num">2023.01.class="num">01 class="num">00:class="num">00, with actual being: class="num">55544.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">29.032 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">143198.91357580, on: class="num">2023.02.class="num">01 class="num">00:class="num">00, with actual being: class="num">124659.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">29.151 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">24327.24201304, on: class="num">2023.03.class="num">01 class="num">00:class="num">00, with actual being: class="num">48737.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">29.267 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">167059.32862279, on: class="num">2023.04.class="num">01 class="num">00:class="num">00, with actual being: class="num">69888.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">29.357 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">67301.73632062, on: class="num">2023.05.class="num">01 class="num">00:class="num">00, with actual being: class="num">74073.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">29.470 ct_15_w_n(GER30,MN1) class="type">void OnStart() Objects forecast is: class="num">87406.10063493, on: class="num">2023.06.class="num">01 class="num">00:class="num">00, with actual being: class="num">104992.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">29.470 ct_15_w_n(GER30,MN1) class="type">void OnStart() correlation is: class="num">0.02260757 with morph constant as: class="num">0.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">45.341 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">45860.00000000, on: class="num">2020.01.class="num">01 class="num">00:class="num">00, with actual being: class="num">109780.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">45.381 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">109780.00000000, on: class="num">2020.02.class="num">01 class="num">00:class="num">00, with actual being: class="num">210420.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">45.420 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">210420.00000000, on: class="num">2020.03.class="num">01 class="num">00:class="num">00, with actual being: class="num">87250.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">45.466 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">87250.00000000, on: class="num">2020.04.class="num">01 class="num">00:class="num">00, with actual being: class="num">58380.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">45.508 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">1581489.63705309, on: class="num">2020.05.class="num">01 class="num">00:class="num">00, with actual being: class="num">58370.00000000 ... class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">46.685 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">153576.16806574, on: class="num">2023.01.class="num">01 class="num">00:class="num">00, with actual being: class="num">30453.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">46.710 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">222058.94934174, on: class="num">2023.02.class="num">01 class="num">00:class="num">00, with actual being: class="num">77051.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">46.739 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">50993.02046828, on: class="num">2023.03.class="num">01 class="num">00:class="num">00, with actual being: class="num">30780.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">46.784 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">66795.26430874, on: class="num">2023.04.class="num">01 class="num">00:class="num">00, with actual being: class="num">45453.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">46.832 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">75492.86146563, on: class="num">2023.05.class="num">01 class="num">00:class="num">00, with actual being: class="num">28845.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">46.876 ct_15_w_n(UK100,MN1) class="type">void OnStart() Objects forecast is: class="num">92298.50964858, on: class="num">2023.06.class="num">01 class="num">00:class="num">00, with actual being: class="num">49472.00000000 class="num">2023.07.class="num">30 class="num">15:class="num">12:class="num">46.876 ct_15_w_n(UK100,MN1) class="type">void OnStart() correlation is: class="num">0.14415363 with morph constant as: class="num">0.00000000
◍ 把这条线请下神坛
用财经日历构图、把波动率曲面当范畴、再用滞后标普500的函子关系去推下一步波动，这套框架本质上只是把主观判断转成了可装配的映射。前几篇留的 ct_15_w.mq5（17.23 KB）和 ct_15_w_n.mq5（18.41 KB）已经把尾随止损的挪移写死在向导里，读者照着自己策略改阈值比照搬更实际。 我们一直用的增量近似是线性或二次关系，域到协域的粗糙映射能跑通，但人工神经网络定义同一映射在下一篇才展开，不在这收口。外汇与贵金属杠杆高、滑点突兀，函子推出来的信号只调仓位和止损，别当入场圣旨。 至此变形与函子两块拼图凑齐，下次直接拿网络替掉手调增量，看映射是不是更贴真实波动。
把图到波动的映射交给小布
这些把 MQL5 日历重构为图、再算函子配对标普波动的诊断，小布盯盘的 AIGC 已内置，打开对应品种页即可看到预计算的范畴映射，你只管判断信号可信度。
常见问题
线性序对象只能顺次连一个邻点，图论允许顶点连多个对象并形成环，更适合表达财经日历里零售销售、CPI、PMI 互导的网状关系。
标普按时间逐点排列，而图的多行顶点对被捆绑进单个波动数据点，域元素成组映射，结构不对等故只可能是函子而非同构。
小布已内置日历转图与函子配对标普波动的模块，不用自己写 MQL5 抓取与重构，打开品种页直接读映射结果即可。
本文不验证该环准确性，只示范构图法；真实交易前需用历史数据回测环内传导倾向，外汇贵金属高风险，结论仅作概率参考。
将图范畴顶点对行数与标普波动点配对，观察多新闻共振时的波动倾斜，具体实现见本篇第四节对交易者的影响与收束说明。