⛏️

攫取盈利至最后的点位·进阶篇

（2/3）· 跳过柱线指标的红海，用 MT5 测试器当挖掘机在报价流里捞宝藏

案例拆解新手友好第 2/3 篇

多数人找交易系统只在历史柱线和指标里打转，那片土壤早被翻烂了。把即时报价当成更接近原始的矿层，才有可能碰到还没被大众踩过的形态。本文接上篇，继续把广域搜索的铲子磨快。

本章目录

手工筛 81 个变体，先看一段夜间收益
时段切分后的回测表现差异
EURCHF 的佣金吞噬与网格幻象
把执行细节变成真金白银

手工筛 81 个变体，先看一段夜间收益

个品种的全量计算在 10 小时内跑完，MT5 测试器多核并行顺带把 CPU 烤了 10 小时暖气。优化缓存落库后，我手工翻了 81 个变体——这步本可脚本化，但还没写，所以实打实耗了一小时。

挑一个品种举例：按 BestInterval 条件排序，自定义品种不限制负余额，所以余额绝对值不用管，只看成交密度。红色列里选一条记录做单测，交易图显示是全天候跑的。单测末尾日志写明：想拿 4035 点利润，得把交易时段卡在晚 8 点到早 8 点。进 EA 输入参数把 BestInterval 激活，确认 MT5 关掉优化模式再跑，日志里成交密集、余额曲线平滑，看起来不错。但凡图表在优化间隔外崩掉的，基本是数理性过度拟合——实盘大概率亏光。MT5 的前驱测试我视为自欺，更靠谱的做法是拿整个历史跑两三个好结果，全历史不行就弃用，前驱测试多跑也只是过度拟合的副产品。黄金这类品种常出现全历史优化值拉胯的情况，红框是优化间隔；测试周期远长于优化间隔时，稳定性才上来。下面这段日志是夜间时段（20:00–08:00，11 小时）开启 BestInterval 的实跑摘要。

MQL5 / C++

BestInterval Action(<span class="keyword">true</span> - single pass &amp; MT4-style &amp; Virtual <span class="keyword">is</span> required) = <span class="keyword">false</span>
Profit = -<span class="number">class="num">2392.00</span> = -<span class="number">class="num">2392.00</span> + <span class="number">class="num">0.00</span> (<span class="number">class="num">0.00</span>%) - Amount of Delete Intervals = <span class="number">class="num">0</span> (<span class="number">class="num">2019.04</span>.<span class="number">class="num">01</span> - <span class="number">class="num">2019.07</span>.<span class="number">class="num">20</span>)
<span class="number">class="num">00</span>:<span class="number">class="num">00</span>:<span class="number">class="num">00</span> - <span class="number">class="num">23</span>:<span class="number">class="num">59</span>:<span class="number">class="num">59</span> : Profit = -<span class="number">class="num">2392.00</span> (<span class="number">class="num">100.00</span>%), Total = <span class="number">class="num">2612</span> (<span class="number">class="num">70.64</span>%), PF = <span class="number">class="num">0.94</span>, Mean = -<span class="number">class="num">0.92</span>, DD = <span class="number">class="num">3840.00</span>, RF = -<span class="number">class="num">0.62</span>
SUMMARY: <span class="number">class="num">00</span>:<span class="number">class="num">00</span>:<span class="number">class="num">00</span> - <span class="number">class="num">23</span>:<span class="number">class="num">59</span>:<span class="number">class="num">59</span> : Profit = -<span class="number">class="num">2392.00</span> (<span class="number">class="num">100.00</span>%), Total = <span class="number">class="num">2612</span> (<span class="number">class="num">70.64</span>%), PF = <span class="number">class="num">0.94</span>, Mean = -<span class="number">class="num">0.92</span>, DD = <span class="number">class="num">3840.00</span>, RF = -<span class="number">class="num">0.62</span>
Profit = <span class="number">class="num">4035.00</span> = -<span class="number">class="num">2392.00</span> + <span class="number">class="num">6427.00</span> (-<span class="number">class="num">268.69</span>%) - Amount of Delete Intervals = <span class="number">class="num">1</span> (<span class="number">class="num">2019.04</span>.<span class="number">class="num">01</span> - <span class="number">class="num">2019.07</span>.<span class="number">class="num">20</span>), <span class="number">class="num">20</span>:<span class="number">class="num">00</span> - <span class="number">class="num">08</span>:<span class="number">class="num">00</span>, CountHours = <span class="number">class="num">11</span>
<span class="number">class="num">00</span>:<span class="number">class="num">00</span>:<span class="number">class="num">00</span> - <span class="number">class="num">07</span>:<span class="number">class="num">58</span>:<span class="number">class="num">54</span> : Profit = <span class="number">class="num">1074.00</span> (<span class="number">class="num">26.62</span>%), Total = <span class="number">class="num">349</span> (<span class="number">class="num">76.22</span>%), PF = <span class="number">class="num">1.21</span>, Mean = <span class="number">class="num">3.08</span>, DD = <span class="number">class="num">709.00</span>, RF = <span class="number">class="num">1.51</span>
<span class="number">class="num">19</span>:<span class="number">class="num">41</span>:<span class="number">class="num">38</span> - <span class="number">class="num">23</span>:<span class="number">class="num">59</span>:<span class="number">class="num">59</span> : Profit = <span class="number">class="num">2961.00</span> (<span class="number">class="num">73.38</span>%), Total = <span class="number">class="num">348</span> (<span class="number">class="num">76.44</span>%), PF = <span class="number">class="num">1.94</span>, Mean = <span class="number">class="num">8.51</span>, DD = <span class="number">class="num">358.00</span>, RF = <span class="number">class="num">8.27</span>
SUMMARY: <span class="number">class="num">00</span>:<span class="number">class="num">00</span>:<span class="number">class="num">00</span> - <span class="number">class="num">23</span>:<span class="number">class="num">59</span>:<span class="number">class="num">59</span> : Profit = <span class="number">class="num">4035.00</span> (<span class="number">class="num">100.00</span>%), Total = <span class="number">class="num">697</span> (<span class="number">class="num">76.33</span>%), PF = <span class="number">class="num">1.49</span>, Mean = <span class="number">class="num">5.79</span>, DD = <span class="number">class="num">484.00</span>, RF = <span class="number">class="num">8.34</span>
BestInterval <span class="keyword">is</span> saved <span class="keyword">in</span> <span class="class="type">class="kw">string">"TesterEA"</span>-file <span class="keyword">in</span> common(MT5)/<span class="keyword">base</span>(MT4) folder.
final balance - InitBalance(<span class="number">class="num">10000.00</span>) + Profit(-<span class="number">class="num">2392.00</span>) without BestInterval.
<span style="background-class="type">class="kw">color:rgb(class="num">255, class="num">246, class="num">200);">OnTester - Profit(<span class="number">class="num">4035.00</span>) with BestInterval.</span>
final balance <span class="number">class="num">7608.00</span> USD
OnTester result <span class="number">class="num">4035</span>
BestInterval Action(<span class="macro">true</span> - single pass &amp; MT4-style &amp; Virtual is required) = <span class="macro">true</span>
Calculation time activated intervals is <span class="number">class="num">2019.07</span>.<span class="number">class="num">23</span> <span class="number">class="num">16</span>:<span class="number">class="num">27</span>:<span class="number">class="num">25</span> - TesterEA(common folder) <span class="number">class="num">00</span>:<span class="number">class="num">13</span>:<span class="number">class="num">14</span> ago.
Amount of Delete Intervals = <span class="number">class="num">1</span> (<span class="number">class="num">2019.04</span>.<span class="number">class="num">01</span> - <span class="number">class="num">2019.07</span>.<span class="number">class="num">20</span>), <span class="number">class="num">20</span>:<span class="number">class="num">00</span> - <span class="number">class="num">08</span>:<span class="number">class="num">00</span>, CountHours = <span class="number">class="num">11</span>

◍ 时段切分后的回测表现差异

把全天按 00:00:00–07:58:54 与 19:41:38–23:59:59 两段拆开看，同一策略的表现完全不是一回事。前一段 349 笔交易、盈利 1074.00（占 26.62%），盈利因子 PF 只有 1.21，最大回撤 DD 709.00，风险回报比 RF 1.51；后一段 348 笔、盈利 2961.00（占 73.38%），PF 拉到 1.94，DD 降到 358.00，RF 冲到 8.27。合并全天 697 笔、总盈利 4035.00（初始余额 10000.00 的 40.35%），整体 PF 1.49、均值盈利 5.79。注意这段统计的 Profit 是按 TickValue=1 且不含佣金与掉期算的，外汇与贵金属实盘加上点差和隔夜费后数值会明显缩水，属高风险验证项。 OnTester 里还有一条对照：不带 BestInterval 优选时段时，虚拟余额 10000.00 加 Profit -2403.00，最终 OnTester 返回 7597。说明优选时段开关直接改变了回测的退出值，想复现的话进 MT5 策略测试器把周期切片跑一遍就能比对。

MQL5 / C++

class="num">00:class="num">00:class="num">00 - class="num">07:class="num">58:class="num">54 : Profit = class="num">1074.00 (class="num">26.62%), Total = class="num">349 (class="num">76.22%), PF = class="num">1.21, Mean = class="num">3.08, DD = class="num">709.00, RF = class="num">1.51
class="num">19:class="num">41:class="num">38 - class="num">23:class="num">59:class="num">59 : Profit = class="num">2961.00 (class="num">73.38%), Total = class="num">348 (class="num">76.44%), PF = class="num">1.94, Mean = class="num">8.51, DD = class="num">358.00, RF = class="num">8.27
SUMMARY: class="num">00:class="num">00:class="num">00 - class="num">23:class="num">59:class="num">59 : Profit = class="num">4035.00 (class="num">100.00%), Total = class="num">697 (class="num">76.33%), PF = class="num">1.49, Mean = class="num">5.79
final balance - InitBalance(class="num">10000.00) + Profit(class="num">4035.00) with BestInterval.
OnTester - Virtual InitBalance(class="num">10000.00) + Profit(-class="num">2403.00) without BestInterval. Profit is calculated with TickValue=class="num">1 and w/o Commission+Swap.
final balance class="num">14035.00 USD
OnTester result class="num">7597

「EURCHF 的佣金吞噬与网格幻象」

多轮测试跑完，品种筛选结果指向 EURCHF。其回测报告里以点为单位的数学期望被标红，这是核心参考量——但该品种来回佣金约 4.40 点，意味着经纪商抽走利润的三分之二，成本占比过高，实盘前必须重新评估点值结构。为剔除非系统数据造成的优化失真，删掉 BestInterval，改挂交易时间输入，从最大回撤结束位（图上绿标）起跑优化；参数多就跑几轮遗传优化。源码里加一段时间过滤，未过滤品种的数学期望收益因此翻了两倍。我也试过网格与马丁类手法，数学收益曲线一度走高却讲不清来源。后来发现 MT5 测试器对净持结算账户用了另一套算法，显示值并非真实策略产出。下方代码重写了 OnTester 的正确期望 payoff 计算。重写后结论很干脆：网格无法改善已优化策略的数学期望，只是改变了账户计算口径。外汇与贵金属杠杆高，这类回测结论仅代表历史样本概率倾向，实盘仍需小资金验证。

MQL5 / C++

class="kw">const class="type">bool TradeTime = (TimeCurrent() < D&class="macro">#x27;class="num">2018.02.class="num">10&class="macro">#x27;) || (TimeCurrent() >= D&class="macro">#x27;class="num">2018.02.class="num">12&class="macro">#x27;);
class=class="str">"cmt">// Calculating expected payoff on hedging accounts corresponds to traditional calculation.
class="type">class="kw">double OnTester()
{
  class="type">class="kw">double Res = class="num">0;
  
  if (HistorySelect(class="num">0, INT_MAX))
    for (class="type">int i = HistoryDealsTotal() - class="num">1; i >= class="num">0; i--)
      Res += HistoryDealGetDouble(HistoryDealGetTicket(i), DEAL_VOLUME);
  
  Res /= class="num">2;
  
  if (Res)
    Res = TesterStatistics(STAT_PROFIT) / Res;
  
  class="kw">return(Res);
}

把执行细节变成真金白银

选经纪商时别只看点差。要拉出历史 tick 查限价单真实成交方式：有的券商用市价单兜底限价单，允许负数滑点；有的禁止限价挂单部分成交。这两类都该避开——在高频策略里每 0.1 点都直接进数学期望。实盘我跑在 MT4，但回测逻辑来自 MT5 的跨平台库。EA 启动后先通过 HistoryTicks 拉满历史报价，再用虚拟函数库在沙盒里对齐 tick 级环境；设置里勾选“整段交易历史”显示到当前，能把输入参数出错概率压到最低。关 EA 再开，虚拟环境接着跑，像没断过。真实环境持续和这个“完美执行”的虚拟环境同步，目的只有一个：把每笔该吃到的滑点都吃进来。我几乎把所有系统都这么处理，对这种薄利高频尤其关键。只锁一组参数等于赌未来。我直接挑了 8 个优化变体，等资金分摊，八个图表各挂独立 MagicNumber 同时跑。看起来像网格，其实是八套互不相关的系统，谁将来活得好就靠分布扛住不确定性。外汇与贵金属杠杆高，这类多实例跑法仍可能整体回撤，需实时盯保证金。