📘

继续漫步优化（第二部分）：为任意机器人创建优化报告的机制·综合运用

第 3/3 篇

本章目录

逐笔统计里的连盈连亏与回撤锚点
回测报告里的逐笔聚合逻辑
按交易日拆分盈亏与胜率统计
回撤恢复因子与夏普的类图计算落点
夏普式收益波动比的逐行实现
从逐笔盈亏到 VaR 与分布曲线的落地
用概率密度函数给收益分布建模
把报告生成器接进你的优化流水线

逐笔统计里的连盈连亏与回撤锚点

这段逻辑跑在每笔成交闭合之后，核心是把单笔盈亏 pl 按是否标准化为一手（isOneLot）做分流统计。dealsCounter 里同时维护连续盈利计数 Profit 与连续回撤计数 DD，盈利一笔就清零 DD、亏损一笔就清零 Profit，因此 out.dealsCounter.DD 与 Profit 取的是历史最大连败与连胜长度。 R_arr 数组记录的是盈亏序列的「段」：n 为 1 表示盈利、0 表示亏损，仅当上一段结尾值不同时才压入新元素，等价于把 101001 压缩成 10101 的连段标记。回测时直接读这个数组长度分布，就能看出策略是碎步盈利还是偶发大亏。最大单笔回撤与最大单笔利润分别用 Max_DD_forDeal / Max_Profit_forDeal 承接，并保存对应平仓时间 DT_close。Accumulated_DD 只累加负 pl，Accumulated_Profit 只累加正 pl，二者分离便于后续算 Calmar 类比率。

基于累计 PL 的浮盈高点回撤更有意思：maxPL 取历史最高累计收益，当当前 PL 低于该高点时，DD 按 PL-maxPL（盈利回吐）或 -(	PL	+maxPL)（击穿成本线）计算，Max_DD_byPL 存的是全局最差水下偏离。外汇与贵金属杠杆品种中该值可能瞬间超过本金 50%，属高风险信号，需用 MT5 跑历史品种验证阈值。

MQL5 / C++

  class="type">bool isOneLot)
  {
   class="type">class="kw">double pl=(isOneLot ? deal.pl_oneLot : deal.pl_forDeal); class=class="str">"cmt">//PL
   class="type">int n=class="num">0;
class=class="str">"cmt">// Number of profits and losses
   if(pl>=class="num">0)
     {
      out.Total_Profit_numDeals++;
      n=class="num">1;
      out.dealsCounter.Counter.DD=class="num">0;
      out.dealsCounter.Counter.Profit++;
     }
   else
     {
      out.Total_DD_numDeals++;
      out.dealsCounter.Counter.DD++;
      out.dealsCounter.Counter.Profit=class="num">0;
     }
   out.dealsCounter.DD=MathMax(out.dealsCounter.DD,out.dealsCounter.Counter.DD);
   out.dealsCounter.Profit=MathMax(out.dealsCounter.Profit,out.dealsCounter.Counter.Profit);
class=class="str">"cmt">// Series of profits and losses
   class="type">int s=ArraySize(out.R_arr);
   if(!(s>class="num">0 && out.R_arr[s-class="num">1]==n))
     {
      ArrayResize(out.R_arr,s+class="num">1,s+class="num">1);
      out.R_arr[s]=n;
     }
   out.PL+=pl; class=class="str">"cmt">//Total PL
class=class="str">"cmt">// Max Profit / DD
   if(out.Max_DD_forDeal>pl)
     {
      out.Max_DD_forDeal=pl;
      out.DT_Max_DD_forDeal=deal.DT_close;
     }
   if(out.Max_Profit_forDeal<pl)
     {
      out.Max_Profit_forDeal=pl;
      out.DT_Max_Profit_forDeal=deal.DT_close;
     }
class=class="str">"cmt">// Accumulated Profit / DD
   out.Accomulated_DD+=(pl>class="num">0 ? class="num">0 : pl);
   out.Accomulated_Profit+=(pl>class="num">0 ? pl : class="num">0);
class=class="str">"cmt">// Extreme profit values
   class="type">class="kw">double maxPL=MathMax(out.Max_Profit_byPL,out.PL);
   if(compareDouble(maxPL,out.Max_Profit_byPL)==class="num">1)class=class="str">"cmt">/* || !isNot_firstDeal*/)class=class="str">"cmt">// another check is needed to save the date
     {
      out.DT_Max_Profit_byPL=deal.DT_close;
      out.Max_Profit_byPL=maxPL;
     }
   class="type">class="kw">double maxDD=out.Max_DD_byPL;
   class="type">class="kw">double DD=class="num">0;
   if(out.PL>class="num">0)
      DD=out.PL-maxPL;
   else
      DD=-(MathAbs(out.PL)+maxPL);
   maxDD=MathMin(maxDD,DD);
   if(compareDouble(maxDD,out.Max_DD_byPL)==-class="num">1)class=class="str">"cmt">/* || !isNot_firstDeal*/)class=class="str">"cmt">// another check is needed to save the date
     {
      out.Max_DD_byPL=maxDD;
      out.DT_Max_DD_byPL=deal.DT_close;
     }

「回测报告里的逐笔聚合逻辑」

MT5 里做交易报告生成时，单笔成交（Deal）如何被归并进总统计，决定了你看到的 Equity 曲线和因子图是否可信。下面这段 CReportCreator::CalcData 展示了非导出模式下，一笔 deal 被塞进哪些计算通道。注意 isBH 参数：当它为 false，才走完整图表填充。也就是说，如果是某种“仅基础统计”的调用，PL 图、每日盈亏、夏普、Z 分数全部跳过，num_deals 却依然自增——这可能导致你拿到的报告交易数正确，但曲线为空。从代码看，单笔会同时喂给 oneLot（标准化 1 手）和 total（实际手数）两套 CalcData_item，再分别灌入 PL_total / PL_oneLot / PL_Indicative 三条权益线。外汇与贵金属波动大，用 oneLot 归一能横向比策略，但实盘高风险，归一不等于实盘资金曲线。因子图部分覆盖了 Profit Factor、Recovery Factor、WinCoef、Sharpe（代码里拼成 ShartRatio）、Altman Z Score，且 each 都分 OneLot 与 Total 双版本。Z 分数计算直接取 num_deals、R_arr 大小、盈利/亏损笔数，说明它是基于每笔结果序列的游程检验，而非简单均值。

MQL5 / C++

out.DD_percent=(balance>class="num">0 ?(MathAbs(DD)/(maxPL>class="num">0 ? maxPL : balance)) :(maxPL>class="num">0 ?(MathAbs(DD)/maxPL) : class="num">0));
}
class="type">void CReportCreator::CalcData(class="kw">const DealDetales &deal,CalculationData &out,class="type">bool isBH)
  {
   out.num_deals++; class=class="str">"cmt">// Counting the number of deals
   CalcData_item(deal,out.oneLot,true);
   CalcData_item(deal,out.total,false);
   if(!isBH)
     {
      class=class="str">"cmt">// Fill PL graphs
      CalcPL(deal,out,PL.PL_total,_Total);
      CalcPL(deal,out,PL.PL_oneLot,_OneLot);
      CalcPL(deal,out,PL.PL_Indicative,_Indicative);
      class=class="str">"cmt">// Fill PL Histogram graphs
      CalcPLHist(deal,out,PL_hist.PL_total,_Total);
      CalcPLHist(deal,out,PL_hist.PL_oneLot,_OneLot);
      CalcPLHist(deal,out,PL_hist.PL_Indicative,_Indicative);
      class=class="str">"cmt">// Fill PL graphs by days
      CalcDailyPL(DailyPL_data.absolute_close,CALC_FOR_CLOSE,deal);
      CalcDailyPL(DailyPL_data.absolute_open,CALC_FOR_OPEN,deal);
      CalcDailyPL(DailyPL_data.avarage_close,CALC_FOR_CLOSE,deal);
      CalcDailyPL(DailyPL_data.avarage_open,CALC_FOR_OPEN,deal);
      class=class="str">"cmt">// Fill Profit Factor graphs
      ProfitFactor_chart_calc(CoefChart_data.OneLot_ProfitFactor_chart,out,deal,true);
      ProfitFactor_chart_calc(CoefChart_data.Total_ProfitFactor_chart,out,deal,false);
      class=class="str">"cmt">// Fill Recovery Factor graphs
      RecoveryFactor_chart_calc(CoefChart_data.OneLot_RecoveryFactor_chart,out,deal,true);
      RecoveryFactor_chart_calc(CoefChart_data.Total_RecoveryFactor_chart,out,deal,false);
      class=class="str">"cmt">// Fill winning coefficient graphs
      WinCoef_chart_calc(CoefChart_data.OneLot_WinCoef_chart,out,deal,true);
      WinCoef_chart_calc(CoefChart_data.Total_WinCoef_chart,out,deal,false);
      class=class="str">"cmt">// Fill Sharpe Ration graphs
      ShartRatio_chart_calc(CoefChart_data.OneLot_ShartRatio_chart,PL.PL_oneLot,dealclass=class="str">"cmt">/*,out.isNot_firstDeal*/);
      ShartRatio_chart_calc(CoefChart_data.Total_ShartRatio_chart,PL.PL_total,dealclass=class="str">"cmt">/*,out.isNot_firstDeal*/);
      class=class="str">"cmt">// Fill Z Score graphs
      AltmanZScore_chart_calc(CoefChart_data.OneLot_AltmanZScore_chart,(class="type">class="kw">double)out.num_deals,
                  (class="type">class="kw">double)ArraySize(out.oneLot.R_arr),(class="type">class="kw">double)out.oneLot.Total_Profit_numDeals,
                  (class="type">class="kw">double)out.oneLot.Total_DD_numDealsclass=class="str">"cmt">/*,out.isNot_firstDeal*/,deal);
      AltmanZScore_chart_calc(CoefChart_data.Total_AltmanZScore_chart,(class="type">class="kw">double)out.num_deals,
                  (class="type">class="kw">double)ArraySize(out.total.R_arr),(class="type">class="kw">double)out.total.Total_Profit_numDeals,
                  (class="type">class="kw">double)out.total.Total_DD_numDealsclass=class="str">"cmt">/*,out.isNot_firstDeal*/,deal);

◍ 按交易日拆分盈亏与胜率统计

报告生成器里这段逻辑，把每笔成交按星期几归类，分别累加盈利和回撤。 CalcDailyPL 依次调用 cmpDay，对周一到周五各跑一次；cmpDay 内部先判断成交按平仓日还是开仓日算（CALC_FOR_CLOSE 取 day_close，否则取 day_open），命中对应星期才累加。盈利侧：deal.pl_forDeal>0 时，ans.Profit 加该笔盈亏，numTrades_profit 自增；亏损侧：<0 时 ans.Drawdown 加 MathAbs 绝对值，numTrades_drawdown 自增。这样一天内多笔同星期成交会被合并统计。 avarageDay 做均值化：Profit 除以盈利笔数、Drawdown 除以亏损笔数，若某侧笔数为 0 则不除避免 NaN。 ProfitFactor_chart_calc 取累计盈利与累计回撤绝对值，按 isOneLot 切换单手或总仓口径，写入 CoefChart_item 的时间序列，供后续绘 PF 曲线。外汇与贵金属品种波动剧烈，周统计仅反映历史分布，实盘样本不足时结论倾向失真，需谨慎。

MQL5 / C++

else class=class="str">"cmt">// Fill PL Buy and Hold graphs
   {
   CalcPL(deal,out,BH.PL_total,_Total);
   CalcPL(deal,out,BH.PL_oneLot,_OneLot);
   CalcPL(deal,out,BH.PL_Indicative,_Indicative);
   CalcPLHist(deal,out,BH_hist.PL_total,_Total);
   CalcPLHist(deal,out,BH_hist.PL_oneLot,_OneLot);
   CalcPLHist(deal,out,BH_hist.PL_Indicative,_Indicative);
   }
if(!out.isNot_firstDeal)
   out.isNot_firstDeal=true; class=class="str">"cmt">// Flag "It is NOT the first deal"
}
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Create a structure of trading during a day                       |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">void CReportCreator::CalcDailyPL(DailyPL &out,DailyPL_calcBy calcBy,class="kw">const DealDetales &deal)
  {
  cmpDay(deal,MONDAY,out.Mn,calcBy);
  cmpDay(deal,TUESDAY,out.Tu,calcBy);
  cmpDay(deal,WEDNESDAY,out.We,calcBy);
  cmpDay(deal,THURSDAY,out.Th,calcBy);
  cmpDay(deal,FRIDAY,out.Fr,calcBy);
  }
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Save resulting PL/DD for the day                                |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">void CReportCreator::cmpDay(class="kw">const DealDetales &deal,ENUM_DAY_OF_WEEK etalone,PLDrawdown &ans,DailyPL_calcBy calcBy)
  {
  ENUM_DAY_OF_WEEK day=(calcBy==CALC_FOR_CLOSE ? deal.day_close : deal.day_open);
  if(day==etalone)
    {
    if(deal.pl_forDeal>class="num">0)
      {
      ans.Profit+=deal.pl_forDeal;
      ans.numTrades_profit++;
      }
    else
      if(deal.pl_forDeal<class="num">0)
        {
        ans.Drawdown+=MathAbs(deal.pl_forDeal);
        ans.numTrades_drawdown++;
        }
    }
  }
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Average resulting PL/DD for the day                             |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">void CReportCreator::avarageDay(PLDrawdown &day)
  {
  if(day.numTrades_profit>class="num">0)
    day.Profit/=day.numTrades_profit;
  if(day.numTrades_drawdown > class="num">0)
    day.Drawdown/=day.numTrades_drawdown;
  }
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Calculate Profit Factor                                          |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">void CReportCreator::ProfitFactor_chart_calc(CoefChart_item &out[],CalculationData &data,class="kw">const DealDetales &deal,class="type">bool isOneLot)
  {
  CoefChart_item item;
  item.DT=deal.DT_close;
  class="type">class="kw">double profit=(isOneLot ? data.oneLot.Accomulated_Profit : data.total.Accomulated_Profit);
  class="type">class="kw">double dd=MathAbs(isOneLot ? data.oneLot.Accomulated_DD : data.total.Accomulated_DD);

回撤恢复因子与夏普的类图计算落点

恢复因子（Recovery Factor）衡量累计盈亏相对最大回撤的覆盖能力。代码里当 dd==0 时直接把 coef 置 0，注释写明理想情况应为正无穷，实盘里这意味着账户从未回撤，概率极低，外汇与贵金属品种的高波动特性下更几乎不可能出现。盈利/回撤比那行用 pl/dd 计算，pl 按 isOneLot 切换单标准手与总仓口径，dd 取 MathAbs 后的最大回撤绝对值。若你跑 EURUSD 的 M15 回测，某段样本 dd 为 120 点、pl 为 360 点，恢复因子就是 3.0，代表盈利是回撤的 3 倍。胜率系数另走一条路：用 (profit/n_profit)/(dd/n_dd) 算每笔盈利单均值对每笔回撤单均值的比，n_profit 或 n_dd 为 0 时 coef 赋 0，避免除零。夏普类函数 ShartRatio_calc 要求样本数 total>=2，从 i=1 起循环且跳过上一笔 Profit 为 0 的节点，说明它基于非零盈亏步长估计序列波动。这几个函数在 MT5 里接在 CReportCreator 的报告流水线中，改 coef 公式前先确认 out[] 的 ArrayResize 步长 s+1 是否和你自己的图表点数匹配，否则容易出现尾部 item 丢失。

MQL5 / C++

if(dd==class="num">0)
   item.coef=class="num">0;
else
   item.coef=profit/dd;
class="type">int s=ArraySize(out);
ArrayResize(out,s+class="num">1,s+class="num">1);
out[s]=item;
}
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Calculate Recovery Factor                                      |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">void CReportCreator::RecoveryFactor_chart_calc(CoefChart_item &out[],CalculationData &data,class="kw">const DealDetales &deal,class="type">bool isOneLot)
  {
  CoefChart_item item;
  item.DT=deal.DT_close;
  class="type">class="kw">double pl=(isOneLot ? data.oneLot.PL : data.total.PL);
  class="type">class="kw">double dd=MathAbs(isOneLot ? data.oneLot.Max_DD_byPL : data.total.Max_DD_byPL);
  if(dd==class="num">0)
    item.coef=class="num">0;class=class="str">"cmt">//ideally it should be plus infinity
  else
    item.coef=pl/dd;
  class="type">int s=ArraySize(out);
  ArrayResize(out,s+class="num">1,s+class="num">1);
  out[s]=item;
  }
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Calculate Win Rate                                             |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">void CReportCreator::WinCoef_chart_calc(CoefChart_item &out[],CalculationData &data,class="kw">const DealDetales &deal,class="type">bool isOneLot)
  {
  CoefChart_item item;
  item.DT=deal.DT_close;
  class="type">class="kw">double profit=(isOneLot ? data.oneLot.Accomulated_Profit : data.total.Accomulated_Profit);
  class="type">class="kw">double dd=MathAbs(isOneLot ? data.oneLot.Accomulated_DD : data.total.Accomulated_DD);
  class="type">int n_profit=(isOneLot ? data.oneLot.Total_Profit_numDeals : data.total.Total_Profit_numDeals);
  class="type">int n_dd=(isOneLot ? data.oneLot.Total_DD_numDeals : data.total.Total_DD_numDeals);
  if(n_dd == class="num">0 || n_profit == class="num">0)
    item.coef = class="num">0;
  else
    item.coef=(profit/n_profit)/(dd/n_dd);
  class="type">int s=ArraySize(out);
  ArrayResize(out,s+class="num">1,s+class="num">1);
  out[s]=item;
  }
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Calculate Sharpe Ratio                                         |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">class="kw">double CReportCreator::ShartRatio_calc(PLChart_item &data[])
  {
  class="type">int total=ArraySize(data);
  class="type">class="kw">double ans=class="num">0;
  if(total>=class="num">2)
    {
    class="type">class="kw">double pl_r=class="num">0;
    class="type">int n=class="num">0;
    for(class="type">int i=class="num">1; i<total; i++)
      {
      if(data[i-class="num">1].Profit!=class="num">0)
        {

「夏普式收益波动比的逐行实现」

这段逻辑在算一个类夏普比率：用相邻笔权益的差值增长率均值 pl_r 减去无风险收益 r，再除以增长率序列的样本标准差 std，得出风险调整后收益 ans。注意分母用了 n-1 而不是 n，是典型的无偏样本标准差写法，外汇与贵金属回测里样本量小的时候这个细节会明显影响数值。下面拆一下核心几行：pl_r 累加的是 (data[i].Profit-data[i-1].Profit)/data[i-1].Profit，也就是逐笔权益环比增长率；只有 data[i-1].Profit 不为 0 才计入，避免除零。n>=2 才做平均，否则前面循环白跑。标准差那段把每个增长率与均值的差平方后求和，最后 MathSqrt(std/(double)(n-1)) 开方。ans 在 std 为 0 时直接返回 0，说明策略收益完全无波动时这套指标失效，实盘里概率极低但样本构造不当会触发。 NormalPDF_chart_calc 里顺手算了绝对收益和增长收益的均值与标准差，并清空 distribution 数组准备画正态拟合。Mx_growth 只从 i>0 且前一笔非空开始累加，因此增长类统计的有效样本数 n 通常比 total 少 1，调参时别拿 total 去当分母。

MQL5 / C++

pl_r+=(data[i].Profit-data[i-class="num">1].Profit)/data[i-class="num">1].Profit;
   n++;
   }
  }
  if(n>=class="num">2)
   pl_r/=(class="type">class="kw">double)n;
  class="type">class="kw">double std=class="num">0;
  n=class="num">0;
  for(class="type">int i=class="num">1; i<total; i++)
   {
    if(data[i-class="num">1].Profit!=class="num">0)
     {
      std+=MathPow((data[i].Profit-data[i-class="num">1].Profit)/data[i-class="num">1].Profit-pl_r,class="num">2);
      n++;
     }
   }
  if(n>=class="num">2)
   std=MathSqrt(std/(class="type">class="kw">double)(n-class="num">1));
  ans=(std!=class="num">0 ?(pl_r-r)/std : class="num">0);
   }
  class="kw">return ans;
  }
class="type">void CReportCreator::NormalPDF_chart_calc(DistributionChart &out,PLChart_item &data[])
  {
   class="type">class="kw">double Mx_absolute=class="num">0,Mx_growth=class="num">0,Std_absolute=class="num">0,Std_growth=class="num">0;
   class="type">int total=ArraySize(data);
   ZeroMemory(out.absolute);
   ZeroMemory(out.growth);
   ZeroMemory(out.absolute.VaR);
   ZeroMemory(out.growth.VaR);
   ArrayFree(out.absolute.distribution);
   ArrayFree(out.growth.distribution);
   if(total>=class="num">2)
    {
     class="type">int n=class="num">0;
     for(class="type">int i=class="num">0; i<total; i++)
      {
       Mx_absolute+=data[i].Profit;
       if(i>class="num">0 && data[i-class="num">1].Profit!=class="num">0)
        {
         Mx_growth+=(data[i].Profit-data[i-class="num">1].Profit)/data[i-class="num">1].Profit;
         n++;
        }
      }
     Mx_absolute/=(class="type">class="kw">double)total;
     if(n>=class="num">2)
      Mx_growth/=(class="type">class="kw">double)n;
     n=class="num">0;
     for(class="type">int i=class="num">0; i<total; i++)
      {
       Std_absolute+=MathPow(data[i].Profit-Mx_absolute,class="num">2);

◍ 从逐笔盈亏到 VaR 与分布曲线的落地

这段逻辑干两件事：先把逐笔绝对盈亏和环比增长率的标准差算出来，再据此填 VaR 并铺出分布点。外汇与贵金属杠杆高，这类统计只描述历史风险轮廓，不代表未来亏损上限。增长率方差只在 i>0 且上一笔 Profit 非零时累加，分子是 (当前Profit-上一笔Profit)/上一笔Profit 减去均值 Mx_growth 的平方；n 计数后，若 n>=2 才用 n-1 做无偏估计开根号。绝对标准差则直接拿 total-1 除完开平方。 VaR 调用分三档分位：Q_90、Q_95、Q_99 分别喂给绝对和增长两套均值/标准差，输出 out.absolute.VaR.VAR_90~99 与 out.growth 对应字段。若你导出的历史交易笔数不足 30，95% 置信 VaR 可能明显低估尾部。分布构造用一个 for 循环把每笔 Profit 及其概率密度 PDF_calc 塞进 item_a，增长项 item_g 在 i>0 且分母非零时才算环比；最后 sorter.Sort 按 x 升序排好，方便小布盯盘直接画 CDF。

MQL5 / C++

if(i>class="num">0 && data[i-class="num">1].Profit!=class="num">0)
  {
  Std_growth+=MathPow((data[i].Profit-data[i-class="num">1].Profit)/data[i-class="num">1].Profit-Mx_growth,class="num">2);
  n++;
  }
}
Std_absolute=MathSqrt(Std_absolute/(class="type">class="kw">double)(total-class="num">1));
if(n>=class="num">2)
  Std_growth=MathSqrt(Std_growth/(class="type">class="kw">double)(n-class="num">1));
class=class="str">"cmt">// Calculate VaR
out.absolute.VaR.Mx=Mx_absolute;
out.absolute.VaR.Std=Std_absolute;
out.absolute.VaR.VAR_90=VaR(Q_90,Mx_absolute,Std_absolute);
out.absolute.VaR.VAR_95=VaR(Q_95,Mx_absolute,Std_absolute);
out.absolute.VaR.VAR_99=VaR(Q_99,Mx_absolute,Std_absolute);
out.growth.VaR.Mx=Mx_growth;
out.growth.VaR.Std=Std_growth;
out.growth.VaR.VAR_90=VaR(Q_90,Mx_growth,Std_growth);
out.growth.VaR.VAR_95=VaR(Q_95,Mx_growth,Std_growth);
out.growth.VaR.VAR_99=VaR(Q_99,Mx_growth,Std_growth);
class=class="str">"cmt">// Calculate distribution
 for(class="type">int i=class="num">0; i<total; i++)
  {
  Chart_item  item_a,item_g;
  ZeroMemory(item_a);
  ZeroMemory(item_g);
  item_a.x=data[i].Profit;
  item_a.y=PDF_calc(Mx_absolute,Std_absolute,data[i].Profit);
  if(i>class="num">0)
    {
    item_g.x=(data[i-class="num">1].Profit != class="num">0 ?(data[i].Profit-data[i-class="num">1].Profit)/data[i-class="num">1].Profit : class="num">0);
    item_g.y=PDF_calc(Mx_growth,Std_growth,item_g.x);
    }
  class="type">int s=ArraySize(out.absolute.distribution);
  ArrayResize(out.absolute.distribution,s+class="num">1,s+class="num">1);
  out.absolute.distribution[s]=item_a;
  s=ArraySize(out.growth.distribution);
  ArrayResize(out.growth.distribution,s+class="num">1,s+class="num">1);
  out.growth.distribution[s]=item_g;
  }
class=class="str">"cmt">// Ascending
sorter.Sort<Chart_item>(out.absolute.distribution,&chartComparer);
sorter.Sort<Chart_item>(out.growth.distribution,&chartComparer);
}
}
class=class="str">"cmt">//+------------------------------------------------------------------+
class=class="str">"cmt">//| Calculate VaR                                                            |
class=class="str">"cmt">//+------------------------------------------------------------------+
class="type">class="kw">double CReportCreator::VaR(class="type">class="kw">double quantile,class="type">class="kw">double Mx,class="type">class="kw">double Std)
  {

用概率密度函数给收益分布建模

在 MT5 报表模块里，收益分布的概率密度可以直接用正态分布公式算。下面这段 CReportCreator::PDF_calc 就是核心：传入均值 Mx、标准差 Std 和观测点 x，返回该点的密度值。当 Std 不为零时，公式走 MathExp(-0.5*((x-Mx)/Std)^2) / (sqrt(2π)*Std)。若 Std 等于 0，函数直接返回 0，避免除零崩溃。实际验证时，把历史成交的盈亏序列算好 Mx 和 Std，再喂给 PDF_calc 扫一遍 x 轴，就能画出你账户的概率密度曲线。外汇和贵金属杠杆高，这条曲线只描述历史样本分布，对未来回撤只能给概率倾向，不是保本依据。

MQL5 / C++

class="type">class="kw">double CReportCreator::PDF_calc(class="type">class="kw">double Mx,class="type">class="kw">double Std,class="type">class="kw">double x)
  {
   if(Std!=class="num">0)
      class="kw">return MathExp(-class="num">0.5*MathPow((x-Mx)/Std,class="num">2))/(MathSqrt(class="num">2*M_PI)*Std);
   else
      class="kw">return class="num">0;
  }

「把报告生成器接进你的优化流水线」

前面几篇把 C# 函数库和自动优化器的逻辑都铺完了，这一节只是把交易报告生成器单独拎出来收口。它依赖早先写好的报告机制，但接口和内部实现已经改过一轮，目前在多次优化和测试流程里跑过，没崩。解压附件里的两个文件夹（CustomGeneric、History manager）到 MQL5/Include 目录即可，里面是 GenericSorter.mqh、DealHistoryGetter.mqh、ReportCreator.mqh 等七个文件，直接 #include 就能调。外汇和贵金属品种波动大、杠杆高，用这套报告回测历史成交时，别把样本期内的漂亮曲线外推成实盘预期，它只反映过去参数下的概率分布。真要验证，新建一个 EA 把 ReportCreator 挂上，跑一遍前行优化，看生成的成交明细和回撤数据是否对得上你手动统计的数字——对不上就说明某层封装悄悄改了手数逻辑。

←

继续漫步优化（第二部分）：为任意机器人创建优化报告的机制·进阶篇

常见问题

把每笔平仓盈亏按顺序排列，统计连续盈利与连续亏损的最长序列；回撤锚点取权益曲线从高点到恢复前的最低值，直接用逐笔盈亏累加即可定位。

以平仓时间为归属日，将当日所有逐笔盈亏求和得日收益，日胜率=盈利日数/总交易日；注意跨日持仓以平仓日计入，避免重复统计。

可以。小布能读取你的回测逐笔数据，自动聚合成交易日统计并输出回撤恢复因子、收益波动比，省去手写类图计算。

用逐笔盈亏序列的样本标准差（除以n-1），均值取算术平均，比值再按周期年化；注意剔除零盈亏的闲置笔数。

先对逐笔盈亏排序求经验分位数（如5%分位即近似VaR），再用直方图或核密度画分布；外汇贵金属波动大，VaR仅作概率参考。

继续漫步优化（第二部分）：为任意机器人创建优化报告的机制·综合运用

逐笔统计里的连盈连亏与回撤锚点

「回测报告里的逐笔聚合逻辑」

◍ 按交易日拆分盈亏与胜率统计

回撤恢复因子与夏普的类图计算落点

「夏普式收益波动比的逐行实现」

◍ 从逐笔盈亏到 VaR 与分布曲线的落地

用概率密度函数给收益分布建模

「把报告生成器接进你的优化流水线」

常见问题

相关阅读