[{"data":1,"prerenderedAt":933},["ShallowReactive",2],{"\u002Fzh\u002Fglossary\u002Fwhat-is-curve-fitting":3,"\u002Fzh\u002Fglossary\u002Fwhat-is-curve-fitting-related":241,"\u002Fzh\u002Fglossary\u002Fwhat-is-curve-fitting-langs":930},{"id":4,"title":5,"body":6,"category":224,"date":225,"description":226,"draft":227,"extension":228,"meta":229,"navigation":230,"path":231,"related":232,"seo":236,"stem":237,"term":238,"updated":239,"__hash__":240},"glossary\u002Fzh\u002F5.glossary\u002Fwhat-is-curve-fitting.md","什么是曲线拟合？如何识别过度拟合的策略",{"type":7,"value":8,"toc":216},"minimark",[9,13,17,20,31,34,37,72,75,78,86,89,113,121,129,133,144,179,182,209],[10,11,12],"p",{},"曲线拟合也叫过度拟合（overfitting），指的是反复调整策略参数，直到它贴合测试用的历史数据，连其中的噪音也一并贴合。这样得出的结果把过去描述得分毫不差，对未来却毫无预测力。测试时看起来很出色的系统，一上实盘当月就亏钱，最常见的原因就是它。",[14,15,16],"h2",{"id":16},"工作原理",[10,18,19],{},"每一个可调的输入都是一个自由度。在一张取值网格里搜索，测的就不是一个策略，而是每种组合各做一次实验：",[21,22,27],"pre",{"className":23,"code":25,"language":26},[24],"language-text","实验次数 = v1 × v2 × ... × vk\n\n4 个参数 × 每个 15 个取值 = 50,625 次回测\n50,625 次中的最优者被选中，一部分是因为它的优势，\n一部分是因为它恰好碰上的那些噪音\n","text",[28,29,25],"code",{"__ignoreMap":30},"",[10,32,33],{},"在同一段固定的历史上，实验做得越多，就越可以肯定：赢家的排名是运气给的。测试器不会提这一点，它只把赢家当成唯一的候选报告出来。",[10,35,36],{},"四个信号，大致按可靠程度排列：",[38,39,40,54,60,66],"ul",{},[41,42,43,47,48,53],"li",{},[44,45,46],"strong",{},"参数悬崖","。某个设置往任一方向挪一格，表现就一落千丈。真实的优势会慢慢减弱，拟合出来的优势则是直接掉下悬崖。只在周期 47 时才管用的",[49,50,52],"a",{"href":51},"\u002Fglossary\u002Fwhat-is-a-moving-average","移动平均线","，就是教科书式的例子。",[41,55,56,59],{},[44,57,58],{},"参数太多，样本太少","。一套有六个输入的规则只在 150 笔交易上验证过，就是欠定的。参数和样本之间没有固定比例，但方向毫无争议。",[41,61,62,65],{},[44,63,64],{},"净值曲线平滑得可疑","。真实的收益是参差不齐的。回测里一条近乎笔直的斜线，通常说明要么参数组是拟合出来的，要么系统是靠死扛亏损单活下来的。",[41,67,68,71],{},[44,69,70],{},"样本外崩溃","。在拟合窗口里成立，一出窗口就消失。这是决定性的检验，但只在您第一次看样本外数据时才算数。",[14,73,74],{"id":74},"为什么重要",[10,76,77],{},"曲线拟合的系统会以一种特定而又昂贵的方式失败：它不会一开始就慢慢亏。只要市场和拟合期相似，它就表现得还过得去；一旦市场状态变了，它就崩了，而且往往正是在重仓的时候。交易者会归咎于运气不好，殊不知模型只是在做它被造出来要做的事。",[10,79,80,81,85],{},"解药不是更聪明的测试，而是一份实盘记录：真实承担点差、佣金和",[49,82,84],{"href":83},"\u002Fglossary\u002Fwhat-is-slippage","滑点","，而且时间够长，经历过开发者从没见过的行情。",[14,87,88],{"id":88},"数据显示了什么",[10,90,91,92,96,97,100,101,96,105,108,109,112],{},"以 ShowMyTrades 上有交易历史的公开账户为样本（2026 年 8 月）：",[49,93,95],{"href":94},"\u002Fglossary\u002Fwhat-is-a-profit-factor","盈利因子","中位数 ",[44,98,99],{},"1.28","，历史最深",[49,102,104],{"href":103},"\u002Fglossary\u002Fwhat-is-drawdown","回撤",[44,106,107],{},"9.7%","，每个账户已平仓交易数中位数 ",[44,110,111],{},"171 笔","。",[10,114,115,116,120],{},"这三个数字构成了一个真实可信的参照分布。38.2% 的账户回撤曾超过 20%，17.6% 超过 50%。如果一个策略宣传盈利因子高于 3.0、",[49,117,119],{"href":118},"\u002Fglossary\u002Fwhat-is-maximum-drawdown","最大回撤","只有 4%，它宣称的不是本事，而是自己站在一个由 1,540 万笔已同步交易构成的分布之外。这种说法可能成立，但需要证据，而证据要看实盘。",[10,122,123,124,128],{},"171 笔已平仓交易这个中位数，值得单独记一下：要验证任何东西，这都是",[49,125,127],{"href":126},"\u002Fguides\u002Fhow-much-history-a-track-record-needs","一个很小的样本","，而大多数实盘账户恰恰就是凭这么大的样本被评判的。",[14,130,132],{"id":131},"在-showmytrades-上的位置","在 ShowMyTrades 上的位置",[10,134,135,136,139,140,143],{},"图表模块里的",[44,137,138],{},"逐笔交易增长","图按交易序号而不是日期来画曲线，长期闲置带来的美化效果就此去掉，账户是还在运作，还是只是还开着，一目了然。",[44,141,142],{},"月度收益表","会列出每一个亏损月和每一段“无交易活动”空档：策略在糟糕的月份之后被悄悄重新优化，通常会在那里留下一段明显的停顿。",[10,145,146,147,149,150,156,157,160,161,164,165,149,168,171,172,178],{},"在高级统计模块中，把",[44,148,95],{},"和",[49,151,153],{"href":152},"\u002Fglossary\u002Fwhat-is-standard-deviation-in-trading",[44,154,155],{},"标准差","、",[44,158,159],{},"Z 值（概率）"," 放在一起看：盈利因子很高，单笔收益却异常均匀，说明账户平滑得不像经历过各种行情，而围绕某一段时期调出来的参数组，往往就是这种表现。",[44,162,163],{},"自定义分析","面板里的",[44,166,167],{},"魔术号",[44,169,170],{},"注释筛选","，可以把多策略账户上的某一个 EA（智能交易系统）单独拎出来，只看它的统计数据。",[49,173,175],{"href":174},"\u002Fglossary\u002Fwhat-is-a-verified-track-record",[44,176,177],{},"业绩记录已验证","徽章目前有 65 个账户持有，它证明数据来自经纪商，而不是来自某份报告文件。",[14,180,181],{"id":181},"常见误解",[38,183,184,191,197,203],{},[41,185,186,187,190],{},"“",[44,188,189],{},"它通过了样本外测试。","”只要您看过样本外结果、又据此改动过任何东西，这些数据就变成了样本内数据。机会只有一次。",[41,192,186,193,196],{},[44,194,195],{},"历史越长，越不会过度拟合。","”关键在于观察数量和自由参数的比例，而不是时间跨度。一边加年份一边加过滤条件，反而可能更糟。",[41,198,186,199,202],{},[44,200,201],{},"滚动前向优化（walk-forward）能解决这个问题。","”它能减轻问题，也可能重演问题：在同一个品种上逐个窗口重新优化，搜索的仍然是同一段历史。",[41,204,186,205,208],{},[44,206,207],{},"已经是实盘了，不可能是曲线拟合。","”一出问题就重新调参的系统，是在实盘中被拟合。账户是真的，优势却还在被倒推着编出来。",[10,210,211,212,112],{},"什么样的记录能回答这个问题，什么样的不能，请阅读",[49,213,215],{"href":214},"\u002Fguides\u002Fverified-vs-unverified-track-records","经验证与未经验证的业绩记录",{"title":30,"searchDepth":217,"depth":217,"links":218},2,[219,220,221,222,223],{"id":16,"depth":217,"text":16},{"id":74,"depth":217,"text":74},{"id":88,"depth":217,"text":88},{"id":131,"depth":217,"text":132},{"id":181,"depth":217,"text":181},"策略","2026-08-19T00:00:00.000Z","曲线拟合就是把策略反复优化，直到它把过去的数据描述得完美无缺，却什么也预测不了。有哪些警示信号，为什么只有实盘结果才能下定论。",false,"md",{},true,"\u002Fzh\u002Fglossary\u002Fwhat-is-curve-fitting",[233,234,235],"what-is-backtesting","what-is-a-profit-factor","what-is-a-verified-track-record",{"title":5,"description":226},"zh\u002F5.glossary\u002Fwhat-is-curve-fitting","曲线拟合 (Curve Fitting)","2026-09-10T00:00:00.000Z","seQQ_6nNRjX5XjiWi7epEXN6cZrnk69LgGFRtAA932E",[242,525,710],{"id":243,"title":244,"body":245,"category":513,"date":225,"description":514,"draft":227,"extension":228,"meta":515,"navigation":230,"path":516,"related":517,"seo":521,"stem":522,"term":523,"updated":225,"__hash__":524},"glossary\u002Fzh\u002F5.glossary\u002Fwhat-is-a-profit-factor.md","什么是盈利因子？公式、合理数值与真实数据",{"type":7,"value":246,"toc":506},[247,250,252,258,261,264,270,273,340,342,345,351,358,360,375,382,385,392,403,405,434,461,463,499],[10,248,249],{},"盈利因子就是总盈利除以总亏损：所有盈利交易赚的钱加起来，除以所有亏损交易亏的钱。盈利因子 1.50，意思是账户每亏 $1.00，就赚回 $1.50。低于 1.00，不管胜率多高，账户从结构上就注定亏钱。",[14,251,16],{"id":16},[21,253,256],{"className":254,"code":255,"language":26},[24],"盈利因子 = 总盈利 \u002F |总亏损|\n",[28,257,255],{"__ignoreMap":30},[10,259,260],{},"注意这个公式没有考虑什么：账户交易了多久、用了多少资金、中途跌得有多深。",[10,262,263],{},"它也可以用胜率、平均盈利和平均亏损来写，这样写就能拿来诊断问题：",[21,265,268],{"className":266,"code":267,"language":26},[24],"盈利因子 = (胜率 × 平均盈利) \u002F ((1 − 胜率) × 平均亏损)\n",[28,269,267],{"__ignoreMap":30},[10,271,272],{},"两个截然不同的系统，可能算出同一个数值。胜率 30%、盈利单平均是亏损单的四倍，结果是 1.71。胜率 80%、盈利单只有亏损单的一半，结果是 2.00。两者都能赚钱，但出问题的方式不同，做起来的感受也完全两样。",[274,275,276,288],"table",{},[277,278,279],"thead",{},[280,281,282,285],"tr",{},[283,284,95],"th",{},[283,286,287],{},"解读",[289,290,291,300,308,316,324,332],"tbody",{},[280,292,293,297],{},[294,295,296],"td",{},"低于 1.00",[294,298,299],{},"亏钱，而且是结构性的，不是差一点",[280,301,302,305],{},[294,303,304],{},"1.00 – 1.10",[294,306,307],{},"还在噪声范围内，一周行情不好就没了",[280,309,310,313],{},[294,311,312],{},"1.10 – 1.30",[294,314,315],{},"优势很薄但真实存在，前提是样本够大",[280,317,318,321],{},[294,319,320],{},"1.30 – 2.00",[294,322,323],{},"策略有效",[280,325,326,329],{},[294,327,328],{},"2.00 – 3.00",[294,330,331],{},"很强。请核对交易笔数，以及成本是否算全",[280,333,334,337],{},[294,335,336],{},"高于 3.00",[294,338,339],{},"在长期历史里很少见：通常是样本太短、靠几笔超大盈利撑着，或者还有亏损单没平仓",[14,341,74],{"id":74},[10,343,344],{},"要回答“这个系统到底赚不赚钱”，盈利因子是最干净的单一指标。它也比胜率难包装得多，因为每一笔亏损都按全额计入分母。",[10,346,347,348,350],{},"但它不是风险指标。盈利因子不管顺序：一个账户一个月赚够全部利润、另外十一个月一直失血，和一个每周稳步上涨的账户，数值可以一模一样。所以要和",[49,349,119],{"href":118},"一起看，否则只看到了终点，没看到过程。",[10,352,353,357],{},[49,354,356],{"href":355},"\u002Fglossary\u002Fwhat-is-backtesting","回测","给出的盈利因子同样要小心：优化器追求的就是把这个比值做到最大，所以回测里的高数值，往往反映的是拟合做得多好，而不是策略本身多好。",[14,359,88],{"id":88},[10,361,362,363,366,367,374],{},"以下数字只描述在 ShowMyTrades 上发布的账户，不代表全体交易者。截至 2026 年 8 月，有交易历史的已发布账户中，",[44,364,365],{},"盈利因子中位数为 1.28","，",[44,368,369,373],{},[49,370,372],{"href":371},"\u002Fguides\u002Fwin-rate-is-not-an-edge","胜率中位数","为 68.8%","。两个数字都是扣除成本后的：每笔交易先计入库存费和佣金，再判定是盈利单还是亏损单，所以 1.28 是净值，不是毛值。",[10,376,377],{},[378,379],"img",{"alt":380,"src":381},"ShowMyTrades 公开账户的中位数：68.8% 的胜率与 1.28 的盈利因子并列","\u002Fimages\u002Farticles\u002Fwin-rate-vs-payoff.svg",[10,383,384],{},"同一组数据的两个中位数：胜率看着很高，背后的优势却很薄。",[10,386,387,388,391],{},"要看的正是这两个数放在一起。中位数账户里，将近七成的交易以盈利平仓，可整体优势也只是每亏 $1.00 赚 $1.28。拿一个两项都正好处在中位数的账户，把公式倒推回去：它的平均盈利只有平均亏损的约 ",[44,389,390],{},"0.58 倍","。赢得多和赚得多是两回事，前者要容易制造得多：亏损单扛着不平，盈利单早早落袋，胜率就往上走，盈利因子却往下掉。",[10,393,394,395,398,399,402],{},"同一组账户的夏普比率中位数是 ",[44,396,397],{},"0.05","，中位数账户有 ",[44,400,401],{},"171 笔已平仓交易","。样本这么小，盈利因子高于 3.00 并不能证明系统更强，只能说明样本还太少，最差的那笔交易还没出现。",[14,404,132],{"id":131},[10,406,407,408,411,412,414,415,156,418,156,421,156,424,156,427,149,430,433],{},"每个已发布账户页面的",[44,409,410],{},"高级统计","模块里都有",[44,413,95],{},"，旁边是能解释它的几个数字：",[44,416,417],{},"胜率",[44,419,420],{},"平均盈利",[44,422,423],{},"平均亏损",[44,425,426],{},"期望值",[44,428,429],{},"总交易数",[44,431,432],{},"夏普比率","。这六个数字合起来，能说清单一比值说不清的事。",[10,435,436,437,447,448,149,451,454,455,460],{},"有两个产品细节值得注意。第一，已经算进比值的成本会另外逐项列出：交易表格有",[44,438,439,440,446],{},"盈利（毛额）",[44,441,442,443,149],{},"列，",[44,444,445],{},"库存费","佣金","各占一列；高级统计还汇总了",[44,449,450],{},"佣金合计",[44,452,453],{},"已付库存费合计","，可以看出净值数字里扣掉了多少成本。在所有已发布账户中，这两项合计分别是佣金 $4,782,670、库存费 $862,547。第二，",[49,456,458],{"href":457},"\u002Fguides\u002Fcustom-analysis-and-filter-presets",[44,459,163],{},"可以先筛选（按日期范围、品种、魔术号、方向或手数），再对子集重算整个统计区块。用它可以检验：去掉表现最好的那个季度，或者去掉最赚钱的那个品种，盈利因子是否依然站得住。",[14,462,181],{"id":181},[38,464,465,476,487,493],{},[41,466,186,467,470,471,475],{},[44,468,469],{},"盈利因子高于 1，就说明我在赚钱。","”前提是成本已经算进比值。如果没有，一个毛值 1.05 的高频系统，",[49,472,474],{"href":473},"\u002Fglossary\u002Fwhat-is-cost-percentage","计入佣金和库存费之后","可能持平甚至亏损。",[41,477,186,478,481,482,486],{},[44,479,480],{},"盈利因子高，风险就低。","”它反映不了回撤、仓位大小和盈亏顺序。马丁格尔式逐级加仓的",[49,483,485],{"href":484},"\u002Fglossary\u002Fwhat-is-position-sizing","仓位管理","方案，可以一直显示 4.00，直到哪天突然失灵。",[41,488,186,489,492],{},[44,490,491],{},"不同时间周期之间可以直接比。","”不能，它没有年化。剥头皮交易者 20,000 笔交易得出的 1.20，和波段交易者 60 笔交易得出的 1.20，说的不是一回事。",[41,494,186,495,498],{},[44,496,497],{},"未平仓的交易不影响它。","”会影响，恰恰因为它们不在里面。一直不平的浮动亏损完全不计入总亏损，在平仓之前会一直把比值抬高。",[10,500,501,502,112],{},"该长期追踪哪些指标、按什么顺序看，请参阅",[49,503,505],{"href":504},"\u002Fguides\u002Ftracking-trading-performance","交易表现追踪完全指南",{"title":30,"searchDepth":217,"depth":217,"links":507},[508,509,510,511,512],{"id":16,"depth":217,"text":16},{"id":74,"depth":217,"text":74},{"id":88,"depth":217,"text":88},{"id":131,"depth":217,"text":132},{"id":181,"depth":217,"text":181},"指标","盈利因子是总盈利除以总亏损，低于 1.0 的账户注定亏损。本文给出各数值区间的解读，以及数千个账户的中位数 1.28。",{},"\u002Fzh\u002Fglossary\u002Fwhat-is-a-profit-factor",[518,519,520,233],"what-is-drawdown","what-is-maximum-drawdown","what-is-position-sizing",{"title":244,"description":514},"zh\u002F5.glossary\u002Fwhat-is-a-profit-factor","盈利因子 (Profit Factor)","OLz2UdKZBQ-s3NM0RQISYNe6Jrhtmy0tM84z39S0Thk",{"id":526,"title":527,"body":528,"category":699,"date":225,"description":700,"draft":227,"extension":228,"meta":701,"navigation":230,"path":702,"related":703,"seo":706,"stem":707,"term":708,"updated":225,"__hash__":709},"glossary\u002Fzh\u002F5.glossary\u002Fwhat-is-a-verified-track-record.md","什么是经验证的业绩记录？定义与核查",{"type":7,"value":529,"toc":692},[530,533,535,558,564,567,573,576,578,588,591,593,603,618,620,642,653,655,685],[10,531,532],{},"经验证的业绩记录，是通过只读连接直接从经纪商持续传来的交易结果记录，整条链路上没有任何一个环节能让发布者编辑、删除或调整顺序。这正是“说法”和“证据”的区别。这个词里其实藏着两个不同的问题，关于验证的争论，大多源于把两者混为一谈。",[14,534,16],{"id":16},[10,536,537,538,542,543,547,548,552,553,557],{},"第一个问题：数据是不是真的。这由连接本身来回答：通道只能读取账户，永远不能写入，而且不间断地运行，所以记录是持续积累起来的，而不是挑个方便的时候导出来的。在 ShowMyTrades 上，这条通道是以下四种之一：安装在 MetaTrader 4 或 5 中的 EA（智能交易系统）、使用 ",[49,539,541],{"href":540},"\u002Fglossary\u002Fwhat-is-an-mt4-investor-password","MT4 投资者密码","的直接连接、",[49,544,546],{"href":545},"\u002Fctrader-account-tracking","cTrader OAuth2 授权","，或 ",[49,549,551],{"href":550},"\u002Ftradelocker-tracking","TradeLocker REST API","。这四种方式都不能下单、改单或平仓。只读凭据是行业里实现这一点的通用机制，",[49,554,556],{"href":555},"\u002Fglossary\u002Fwhat-is-myfxbook","Myfxbook"," 等第三方追踪工具也是这样连接 MetaTrader 账户的。",[21,559,562],{"className":560,"code":561,"language":26},[24],"经纪商服务器 → 只读连接 → ShowMyTrades → 公开页面\n　　　　　　   整条链路中没有任何可编辑的环节\n",[28,563,561],{"__ignoreMap":30},[10,565,566],{},"第二个问题：发布页面的人是不是账户的主人。只读数据流能证明交易确实存在，却证明不了交易是谁的，因为投资者密码可以交给任何人。要补上这个缺口，得做一件只有拥有交易权限的人才能做的事：以一个不可能成交的价格挂一笔单，并在注释栏里填上发布者的用户 ID。在 cTrader 上，一次 OAuth 授权就能同时回答这两个问题。",[10,568,569],{},[378,570],{"alt":571,"src":572},"“业绩记录已验证”回答的是：数据是否通过一条无法下单、改单或平仓的通道从经纪商传来；“交易权限已验证”回答的是：发布者是否控制着这个账户","\u002Fimages\u002Farticles\u002Ftwo-badges.svg",[10,574,575],{},"一份记录可能过了一关、没过另一关，所以两者分开统计。",[14,577,74],{"id":74},[10,579,580,581,149,583,587],{},"连接之后的一切都只是算术。只要输入能被改动，盈利因子、",[49,582,104],{"href":103},[49,584,586],{"href":585},"\u002Fglossary\u002Fwhat-is-an-equity-curve","净值曲线","就全是摆设，页面做得再详细也没用。截图、PDF 对账单和电子表格有着同一个毛病：它们都出自请您信任它们的那个人之手。",[10,589,590],{},"连续性和只读访问同样重要。可以暂停、再恢复的记录，能让一个糟糕的月份凭空消失。数据流连续时，中断本身就会在页面上显示出来，这本身就是信息，而不是信息的缺失。",[14,592,88],{"id":88},[10,594,595,596,366,599,602],{},"验证要自己主动去做，还得花点功夫，数字也说明了这一点。截至 2026 年 8 月，在 ShowMyTrades 上发布的 1 万+ 个账户中，只有 ",[44,597,598],{},"65 个带有“业绩记录已验证”",[44,600,601],{},"264 个带有“交易权限已验证”","。证明所有权的账户比证明数据来源的多，因为挂一笔单，比去经纪商那里改密码容易。",[10,604,605,606,610,611,614,615,617],{},"已验证不等于表现好。在有交易历史的公开账户中，",[49,607,609],{"href":608},"\u002Fglossary\u002Fwhat-is-time-weighted-return","时间加权收益率","的中位数为 ",[44,612,613],{},"+3.2%","，63.0% 为正，最深回撤的中位数为 ",[44,616,107],{},"，还有 17.6% 的账户曾经亏掉峰值的一半以上。这个分布来自 15,436,464 笔已同步交易，这才是诚实的记录该有的样子，和广告里的那些完全不同。",[14,619,132],{"id":131},[10,621,622,623,626,627,149,630,633,634,637,638,641],{},"两枚徽章都在每个账户页面头部的",[44,624,625],{},"认证","组里，位于各项数字上方：",[44,628,629],{},"业绩记录",[44,631,632],{},"交易权限","，获得时显示绿色，未获得时显示灰色，每枚都有提示框，说明它在该平台上认证的是什么。同一组里还会显示",[44,635,636],{},"真实账户","或",[44,639,640],{},"模拟账户","，这是根据经纪商服务器判断的，不是用户自己填的。",[10,643,644,645,648,649,652],{},"验证在账户设置中管理，",[44,646,647],{},"投资者密码（可选）"," 字段就在那里：清空它并保存，业绩记录徽章当即变灰。",[44,650,651],{},"完整控制台","小组件也会显示这两枚徽章，因为它嵌入了账户头部，所以放在别人网站上的小组件，也带着自己的数据来源。",[14,654,181],{"id":181},[38,656,657,667,673,679],{},[41,658,659,112,662,666],{},[44,660,661],{},"指向某个追踪网站的链接，不等于验证",[49,663,665],{"href":664},"\u002Fguides\u002Fthird-party-verification-explained","任何平台都能托管未经验证的账户","。要看页面上的徽章，而不是网址里的域名。",[41,668,669,672],{},[44,670,671],{},"“业绩记录已验证”证明不了所有权","。它证明的是交易真实存在。转卖别人投资者密码的人，同样能提供真实的数据流，只是那些交易不是他自己的。",[41,674,675,678],{},[44,676,677],{},"已验证不等于盈利","。徽章只认证数据来源，别的一概不管。之后还得把增长率和回撤放在一起看。",[41,680,681,684],{},[44,682,683],{},"经验证的模拟账户，仍然是模拟账户","。偏偏在实盘成交最吃亏的那些时候，模拟成交显得过于乐观。",[10,686,687,688,112],{},"两枚徽章的验证步骤详解：",[49,689,691],{"href":690},"\u002Fguides\u002Fhow-to-verify-your-account","如何验证您的交易账户",{"title":30,"searchDepth":217,"depth":217,"links":693},[694,695,696,697,698],{"id":16,"depth":217,"text":16},{"id":74,"depth":217,"text":74},{"id":88,"depth":217,"text":88},{"id":131,"depth":217,"text":132},{"id":181,"depth":217,"text":181},"验证","经验证的业绩记录，是通过只读连接直接从经纪商持续获取的业绩数据，发布者无法改动其中任何内容。",{},"\u002Fzh\u002Fglossary\u002Fwhat-is-a-verified-track-record",[704,705,518],"what-is-an-mt4-investor-password","what-is-myfxbook",{"title":527,"description":700},"zh\u002F5.glossary\u002Fwhat-is-a-verified-track-record","经验证的业绩记录 (Verified Track Record)","XXw0XywuXXvGTGPkURwZXnd-tURxBRrjPXvj-rr4h2k",{"id":711,"title":712,"body":713,"category":224,"date":225,"description":920,"draft":227,"extension":228,"meta":921,"navigation":230,"path":922,"related":923,"seo":926,"stem":927,"term":928,"updated":225,"__hash__":929},"glossary\u002Fzh\u002F5.glossary\u002Fwhat-is-backtesting.md","什么是回测？定义、局限与实盘数据",{"type":7,"value":714,"toc":913},[715,718,720,723,729,732,760,768,770,773,776,778,789,807,818,824,827,829,836,864,873,875,905],[10,716,717],{},"回测是用历史价格数据模拟交易策略，估算它在过去会有怎样的表现。一张订单都不用发到经纪商，就能得出净值曲线、盈利因子和回撤。这些结果只是对策略的一种假设，并不是策略的真实记录。",[14,719,16],{"id":16},[10,721,722],{},"测试器逐根回放历史 K 线或 tick，在每一步套用策略的入场和离场规则，记下一次模拟成交。模拟和现实之间的所有差距，都藏在成本模型和成交模型里：",[21,724,727],{"className":725,"code":726,"language":26},[24],"每笔交易净结果 = 毛盈亏 − 点差 − 佣金 − 库存费 − 滑点\n\n点差成本 = 点差（价格单位）× 合约规模 × 手数\n库存费   = 库存费率 × 手数 × 持仓过夜数\n",[28,728,726],{"__ignoreMap":30},[10,730,731],{},"输出靠不靠得住，取决于三项设置：",[38,733,734,740,751],{},[41,735,736,739],{},[44,737,738],{},"建模质量","。MT4 和 MT5 用百分比表示。如果不提供真实 tick 数据，“每个 tick”（Every tick）模式其实是用 1 分钟 K 线插值出来的；“控制点”（control points）和“仅开盘价”（open prices only）都是近似算法，对依赖 K 线内部走势的策略美化得很厉害。",[41,741,742,745,746,750],{},[44,743,744],{},"点差假设","。多数测试器默认用固定点差。实盘点差会在每日展期（rollover）、新闻公布和时段开盘时扩大，而很多系统恰恰在这些时候交易。按 0.8 点建模、实际却要付 4.0 点的",[49,747,749],{"href":748},"\u002Fglossary\u002Fwhat-is-a-spread","点差","，已经是另一个策略了。",[41,752,753,756,757,759],{},[44,754,755],{},"滑点与成交逻辑","。测试器默认订单都按请求价成交。",[49,758,84],{"href":83},"、拒单、重新报价和部分成交，在测试器里统统不存在。",[10,761,762,763,767],{},"还有第四个问题，软件不会告诉您：这组参数本身就是从一轮搜索里筛出来的。如果您跑了 400 个变体，只留下最好的那个，眼前的结果就是事后挑出来的。这就是",[49,764,766],{"href":765},"\u002Fglossary\u002Fwhat-is-curve-fitting","曲线拟合","，也是测试表现和实盘表现之间最大的一道鸿沟。",[14,769,74],{"id":74},[10,771,772],{},"要否决一个坏主意，回测是最便宜的办法；要确认一个好主意，它却是最昂贵的办法。它能说明一套规则在机制上前后一致，扛得住粗略的成本假设，回撤幅度也在您忍受得了的范围内。但它证明不了优势真的存在，因为它用到的每一项输入都是重建出来的，每一个参数都是看过数据之后才定下来的。",[10,774,775],{},"决定资金投向的人，应该把回测当成一个有一票否决权的筛选工具，而永远不要把它当作证据。",[14,777,88],{"id":88},[10,779,780,781,784,785,788],{},"ShowMyTrades 以公开页面发布了 ",[44,782,783],{},"1 万+ 个账户","，背后是 ",[44,786,787],{},"15,436,464 笔已同步的交易","（2026 年 8 月）。没有一笔是模拟的：每一笔都由经纪商成交，再从经纪商那里读取回来。",[10,790,791,792,795,796,798,799,366,801,798,804,806],{},"这些已发布账户绝大多数是自动交易：自动交易占比的中位数为 ",[44,793,794],{},"99%","，53.9% 的账户自动化程度超过 90%，42.2% 低于 10%。自动交易的那部分，主要是投入资金之前做过回测的策略。在全部已发布账户中，",[49,797,95],{"href":94},"中位数为 ",[44,800,99],{},[49,802,432],{"href":803},"\u002Fglossary\u002Fwhat-is-the-sharpe-ratio",[44,805,397],{},"。测试报告里的数字，真正要比的就是这些。",[10,808,809,810,813,814,817],{},"模拟最容易低估的是成本，而这里的成本是实际扣掉的，不是假设出来的：佣金 ",[44,811,812],{},"$4,782,670","，库存费 ",[44,815,816],{},"$862,547","；在产生过库存费的账户中，86.3% 的净库存费为负。策略要隔夜持仓，回测却把库存费设为零，这不是稍微乐观了一点，而是漏掉了一笔反复发生的费用。",[10,819,820],{},[378,821],{"alt":822,"src":823},"点差、佣金和库存费是三项独立的成本：ShowMyTrades 公开账户共被收取 $4,782,670 的佣金和 $862,547 的库存费","\u002Fimages\u002Farticles\u002Fwhere-costs-hide.svg",[10,825,826],{},"测试器用固定点差、又不算库存费，三项成本就漏掉了两项。",[14,828,132],{"id":131},[10,830,831,832,835],{},"回测有自己单独的区域，从不和账户混在一起。您可以上传 MetaTrader 4、MetaTrader 5 或 cTrader 的策略测试器报告，发布成标有 BACKTEST 的页面。页面会检查有没有篡改迹象，并单独列在您交易者主页的一个标签页里。回测页面从不带验证徽章，也不会出现在小组件、探索、投资组合和 Developer API 中，因为模拟结果不是",[49,833,629],{"href":834},"\u002Fglossary\u002Fwhat-is-a-track-record","。账户页面则不同，每一个都是从经纪商只读同步来的，所以上面的数字描述的是真实执行的交易，而不是模拟出来的交易。",[10,837,838,839,149,841,843,844,847,848,851,852,855,856,859,860,863],{},"回测藏起来的东西，这几个模块能揭出来：高级统计模块里的",[44,840,450],{},[44,842,453],{},"，以账户货币显示策略的真实成本；",[44,845,846],{},"终端","面板，报告与交易服务器之间的 ",[44,849,850],{},"Ping"," 和",[44,853,854],{},"重传","，也就是您实际成交时所处的条件；还有",[44,857,858],{},"交易历史与余额变化","，它的",[44,861,862],{},"已平仓交易","标签页按真实成交价列出每一笔已执行的订单，而不是测试器会给的价格。",[10,865,866,868,869,872],{},[44,867,177],{},"徽章（65 个账户持有）和",[44,870,871],{},"交易权限已验证","徽章（264 个账户持有）标出的，是从经纪商到页面这条链路确实经过核查、而不只是自己声称的记录。",[14,874,181],{"id":181},[38,876,877,883,893,899],{},[41,878,186,879,882],{},[44,880,881],{},"它用了每个 tick，所以很准确。","”tick 建模改善的是价格路径，不是成交。经纪商的点差、执行和拒单，仍然没有算进去。",[41,884,186,885,888,889,892],{},[44,886,887],{},"它测试了十年。","”",[49,890,891],{"href":126},"时长不等于独立性","。如果这十年里只有一种占主导的市场状态，那它只是一次观察，而不是 2,500 次。",[41,894,186,895,898],{},[44,896,897],{},"只优化了三个参数。","”三个参数各取二十个值，就是 8,000 次实验。8,000 次实验里的赢家，从构造上就注定有一部分靠的是运气。",[41,900,186,901,904],{},[44,902,903],{},"模拟账户的前向测试已经证实了它。","”模拟服务器的成交偏乐观，也从不拒单。模拟账户的结果只是第二次模拟，算不上第一次实盘测试。",[10,906,907,908,912],{},"无论是测试结果还是实盘数字，在相信它之前，先按",[49,909,911],{"href":910},"\u002Fguides\u002Fverifying-trading-performance-claims","如何验证交易业绩宣传","逐项核对一遍。",{"title":30,"searchDepth":217,"depth":217,"links":914},[915,916,917,918,919],{"id":16,"depth":217,"text":16},{"id":74,"depth":217,"text":74},{"id":88,"depth":217,"text":88},{"id":131,"depth":217,"text":132},{"id":181,"depth":217,"text":181},"回测就是拿历史价格数据模拟交易策略。它能证明什么、证明不了什么，以及为什么与经纪商同步的实盘结果几乎总会和它不一样。",{},"\u002Fzh\u002Fglossary\u002Fwhat-is-backtesting",[924,235,925],"what-is-curve-fitting","what-is-slippage",{"title":712,"description":920},"zh\u002F5.glossary\u002Fwhat-is-backtesting","回测 (Backtesting)","JFA3C0OuDmaFAnqzJS-I02fA4QcPMlLuq0a6Gjsakv8",[931,932],{"path":765},{"path":231},1793898063551]