[{"data":1,"prerenderedAt":834},["ShallowReactive",2],{"\u002Fzh\u002Fglossary\u002Fwhat-is-a-moving-average":3,"\u002Fzh\u002Fglossary\u002Fwhat-is-a-moving-average-related":229,"\u002Fzh\u002Fglossary\u002Fwhat-is-a-moving-average-langs":831},{"id":4,"title":5,"body":6,"category":212,"date":213,"description":214,"draft":215,"extension":216,"meta":217,"navigation":218,"path":219,"related":220,"seo":224,"stem":225,"term":226,"updated":227,"__hash__":228},"glossary\u002Fzh\u002F5.glossary\u002Fwhat-is-a-moving-average.md","什么是移动平均线？SMA 与 EMA 详解",{"type":7,"value":8,"toc":204},"minimark",[9,13,17,20,31,34,37,40,43,52,55,58,84,95,98,102,105,154,157,196],[10,11,12],"p",{},"移动平均线是某个品种最近固定若干周期的平均价格，每走出一根新 K 线就重算一次。它把起伏杂乱的价格变成一条线，跟着市场走，但总是慢半拍。这种滞后不是调调参数就能去掉的缺陷，计算方式本身就决定了它。",[14,15,16],"h2",{"id":16},"工作原理",[10,18,19],{},"两种常见算法的唯一区别，在于怎样给过去的价格分配权重。",[21,22,27],"pre",{"className":23,"code":25,"language":26},[24],"language-text","SMA(n) = (P1 + P2 + ... + Pn) \u002F n\n\nEMA(n): k = 2 \u002F (n + 1)\n        EMA_今日 = (今日价格 × k) + (EMA_昨日 × (1 − k))\n","text",[28,29,25],"code",{"__ignoreMap":30},"",[10,32,33],{},"简单移动平均线（SMA）给每个周期同样的权重，新 K 线一进来，最旧的那根就整根移出窗口。这一点很少有人留意：SMA 拐头，可能只是因为一根老 K 线离开了窗口，跟眼下的行情毫无关系。",[10,35,36],{},"指数移动平均线（EMA）更看重近期价格，旧数据也从不完全丢掉，每根历史 K 线的影响按几何级数递减。所以它对方向变化反应更快，也正因为这一点，假拐点更多。",[10,38,39],{},"滞后可以量化。如果价格沿直线运行，n 周期 SMA 大约落后 (n−1)\u002F2 个周期。日线上的 200 周期均线，从算法上就决定了它反映的是大约一百天前的价格位置。周期缩短，滞后变小，但均线被噪音带着来回掉头的次数也会变多。没有哪个参数能两头兼顾。",[14,41,42],{"id":42},"为什么重要",[10,44,45,46,51],{},"凡是建立在移动平均线上的规则，都带着这份滞后。行情启动后才入场，行情结束后才离场，每一次进出都要付",[47,48,50],"a",{"href":49},"\u002Fglossary\u002Fwhat-is-a-spread","点差","、佣金，可能还有滑点。在趋势行情里，这些成本和吃到的行情相比不算什么；在震荡行情里，结果就全由它们决定：均线来回穿越，每穿一次就是一笔要付费的交易，账户没有哪一笔亏得惊人，却一点点失血。",[10,53,54],{},"所以移动平均线只是一个部件，不是一套策略。它只反映近期的平均价格。仓位多大、止损放哪、眼下的市场适不适合这条规则，它一概不管。",[14,56,57],{"id":57},"数据显示了什么",[10,59,60,61,65,66,69,70,74,75,78,79,83],{},"以 ShowMyTrades 上有交易历史的公开账户为样本（2026 年 8 月），账户已平仓交易数的中位数为 ",[62,63,64],"strong",{},"171 笔","，单笔持仓时间的中位数为 ",[62,67,68],{},"2.4 小时","，",[47,71,73],{"href":72},"\u002Fglossary\u002Fwhat-is-a-profit-factor","盈利因子","的中位数为 ",[62,76,77],{},"1.28","。日内周期上的均线交叉规则正是这个样子：进场频繁，持仓短，超出盈亏平衡的余量很薄，结果往往由",[47,80,82],{"href":81},"\u002Fglossary\u002Fwhat-is-cost-percentage","交易成本","决定。",[10,85,86,87,90,91,94],{},"因此，评估任何均线交叉系统，都要拿交易频率来对照。这些账户总共交易了 ",[62,88,89],{},"1,724,575 手","，这笔交易量产生的佣金达 ",[62,92,93],{},"$4,782,670","。信号触发得越频繁，光是更常做对还不够：每次做对赚到的，还得足以覆盖每次交叉都要交的那笔费用。",[10,96,97],{},"有一点必须如实说明：ShowMyTrades 不会推断某笔交易出自哪个指标或 EA（智能交易系统），所以这些数字都不能单独算到均线系统头上。它们描述的是全部已发布账户这个整体，也只能如实描述这个整体。",[14,99,101],{"id":100},"在-showmytrades-上的位置","在 ShowMyTrades 上的位置",[10,103,104],{},"图表模块画的是净值、增长、盈利和回撤，展示的是账户，而不是价格走势。这里没有指标叠加，因为平台要做的是展示策略实际做了什么，而不是帮您设计策略。",[10,106,107,108,114,115,118,119,122,123,126,127,122,134,126,137,140,141,144,145,148,149,153],{},"您能做的，是检验一个均线系统的真实表现。在",[47,109,111],{"href":110},"\u002Fguides\u002Fcustom-analysis-and-filter-presets",[62,112,113],{},"自定义分析","中按",[62,116,117],{},"魔术号","筛选，就能把共用账户上的某一个 EA 单独拎出来；再用",[62,120,121],{},"交易持续时间（秒）","、",[62,124,125],{},"星期","和",[62,128,129,130,133],{},"交易时段（服务器时间）",[62,131,132],{},"缩小范围，看这条规则实际在什么时候触发。在高级统计中，把","总交易数",[62,135,136],{},"平均持仓时间",[62,138,139],{},"佣金合计","放在一起看，就知道这么高的交易频率赚不赚得回成本。",[62,142,143],{},"分类统计","面板的",[62,146,147],{},"按品种","标签页能看出盈利是不是只来自某个恰好走出趋势的品种，对任何",[47,150,152],{"href":151},"\u002Fglossary\u002Fwhat-is-trend-following","趋势跟踪","规则来说，这都是最关键的问题。",[14,155,156],{"id":156},"常见误解",[158,159,160,173,179,185],"ul",{},[161,162,163,164,167,168,172],"li",{},"“",[62,165,166],{},"交叉就是系统。","”交叉只是一个入场信号。没有仓位规则、离场规则和成本假设，它算不上策略，拿它做的",[47,169,171],{"href":170},"\u002Fglossary\u002Fwhat-is-backtesting","回测","测的也是市场，而不是方法。",[161,174,163,175,178],{},[62,176,177],{},"200 日均线很重要，因为人人都在看。","”看的人多，换来的是关注，不是优势。我们的数据既不能证明、也不能否定任何特定周期，任何诚实的来源也给不了您这样的周期。",[161,180,163,181,184],{},[62,182,183],{},"EMA 更好，因为反应更快。","”反应快，意味着入场更早，假信号也更多。这是取舍，不是升级。",[161,186,163,187,190,191,195],{},[62,188,189],{},"我优化过周期，34 效果最好。","”在已经看过的历史数据上挑出表现最好的周期，这就是",[47,192,194],{"href":193},"\u002Fglossary\u002Fwhat-is-curve-fitting","曲线拟合","，不管最后挑出的数字是多少。",[10,197,198,199,203],{},"想知道一套规则化的系统到底能不能赚回成本，该看哪些指标，请参阅",[47,200,202],{"href":201},"\u002Fguides\u002Ftracking-trading-performance","追踪您的交易表现","。",{"title":30,"searchDepth":205,"depth":205,"links":206},2,[207,208,209,210,211],{"id":16,"depth":205,"text":16},{"id":42,"depth":205,"text":42},{"id":57,"depth":205,"text":57},{"id":100,"depth":205,"text":101},{"id":156,"depth":205,"text":156},"策略","2026-08-19T00:00:00.000Z","移动平均线就是最近 N 个周期的平均价格，每根 K 线更新一次。SMA 和 EMA 各自怎么算，为什么滞后是它的本性，而不是毛病。",false,"md",{},true,"\u002Fzh\u002Fglossary\u002Fwhat-is-a-moving-average",[221,222,223],"what-is-trend-following","what-is-backtesting","what-is-curve-fitting",{"title":5,"description":214},"zh\u002F5.glossary\u002Fwhat-is-a-moving-average","移动平均线 (Moving Average)","2026-09-24T00:00:00.000Z","YIgcHEcTFyB8f6E0MrzklyQbv0SNHGc7yED76RZ3vWM",[230,454,652],{"id":231,"title":232,"body":233,"category":212,"date":213,"description":444,"draft":215,"extension":216,"meta":445,"navigation":218,"path":446,"related":447,"seo":450,"stem":451,"term":452,"updated":213,"__hash__":453},"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":234,"toc":437},[235,238,240,243,249,252,280,286,288,291,294,296,307,327,337,344,347,349,357,386,396,398,429],[10,236,237],{},"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。",[14,239,16],{"id":16},[10,241,242],{},"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：",[21,244,247],{"className":245,"code":246,"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",[28,248,246],{"__ignoreMap":30},[10,250,251],{},"输出靠不靠得住，取决于三项设置：",[158,253,254,260,269],{},[161,255,256,259],{},[62,257,258],{},"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。",[161,261,262,265,266,268],{},[62,263,264],{},"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",[47,267,50],{"href":49},"，已经是另一个策略了。",[161,270,271,274,275,279],{},[62,272,273],{},"滑点与成交逻辑","。测试器默认订单都按请求价成交。",[47,276,278],{"href":277},"\u002Fglossary\u002Fwhat-is-slippage","滑点","、拒单、重新报价和部分成交，在测试器里统统不存在。",[10,281,282,283,285],{},"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",[47,284,194],{"href":193},"，也是测试表现和实盘表现之间最大的一道鸿沟。",[14,287,42],{"id":42},[10,289,290],{},"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。",[10,292,293],{},"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。",[14,295,57],{"id":57},[10,297,298,299,302,303,306],{},"ShowMyTrades 以公开页面发布了 ",[62,300,301],{},"1 万+ 个账户","，背后是 ",[62,304,305],{},"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。",[10,308,309,310,313,314,316,317,69,319,316,323,326],{},"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 ",[62,311,312],{},"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，",[47,315,73],{"href":72},"中位数为 ",[62,318,77],{},[47,320,322],{"href":321},"\u002Fglossary\u002Fwhat-is-the-sharpe-ratio","夏普比率",[62,324,325],{},"0.05","。测试报告里的数字，真正要比的就是这些。",[10,328,329,330,332,333,336],{},"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 ",[62,331,93],{},"，库存费 ",[62,334,335],{},"$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。",[10,338,339],{},[340,341],"img",{"alt":342,"src":343},"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",[10,345,346],{},"测试器用固定点差、又不算库存费，三项成本就漏掉了两项。",[14,348,101],{"id":100},[10,350,351,352,356],{},"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",[47,353,355],{"href":354},"\u002Fglossary\u002Fwhat-is-a-track-record","业绩记录","。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。",[10,358,359,360,126,362,365,366,369,370,373,374,377,378,381,382,385],{},"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",[62,361,139],{},[62,363,364],{},"已付库存费合计","，以账户货币显示策略的真实成本；",[62,367,368],{},"终端","面板，报告与交易服务器之间的 ",[62,371,372],{},"Ping"," 和",[62,375,376],{},"重传","，也就是您实际成交时所处的条件；还有",[62,379,380],{},"交易历史与余额变化","，它的",[62,383,384],{},"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。",[10,387,388,391,392,395],{},[62,389,390],{},"业绩记录已验证","徽章（65 个账户持有）和",[62,393,394],{},"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。",[14,397,156],{"id":156},[158,399,400,406,417,423],{},[161,401,163,402,405],{},[62,403,404],{},"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。",[161,407,163,408,411,412,416],{},[62,409,410],{},"它测试了十年。","”",[47,413,415],{"href":414},"\u002Fguides\u002Fhow-much-history-a-track-record-needs","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。",[161,418,163,419,422],{},[62,420,421],{},"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。",[161,424,163,425,428],{},[62,426,427],{},"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。",[10,430,431,432,436],{},"无论是测试结果还是实盘数字，在相信它之前，先按",[47,433,435],{"href":434},"\u002Fguides\u002Fverifying-trading-performance-claims","如何验证交易业绩宣传","逐项核对一遍。",{"title":30,"searchDepth":205,"depth":205,"links":438},[439,440,441,442,443],{"id":16,"depth":205,"text":16},{"id":42,"depth":205,"text":42},{"id":57,"depth":205,"text":57},{"id":100,"depth":205,"text":101},{"id":156,"depth":205,"text":156},"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",[223,448,449],"what-is-a-verified-track-record","what-is-slippage",{"title":232,"description":444},"zh\u002F5.glossary\u002Fwhat-is-backtesting","回测 (Backtesting)","JFA3C0OuDmaFAnqzJS-I02fA4QcPMlLuq0a6Gjsakv8",{"id":455,"title":456,"body":457,"category":212,"date":213,"description":642,"draft":215,"extension":216,"meta":643,"navigation":218,"path":644,"related":645,"seo":647,"stem":648,"term":649,"updated":650,"__hash__":651},"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":458,"toc":635},[459,462,464,467,473,476,479,510,512,515,521,523,542,550,557,559,570,600,602,628],[10,460,461],{},"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。",[14,463,16],{"id":16},[10,465,466],{},"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：",[21,468,471],{"className":469,"code":470,"language":26},[24],"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",[28,472,470],{"__ignoreMap":30},[10,474,475],{},"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。",[10,477,478],{},"四个信号，大致按可靠程度排列：",[158,480,481,492,498,504],{},[161,482,483,486,487,491],{},[62,484,485],{},"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",[47,488,490],{"href":489},"\u002Fglossary\u002Fwhat-is-a-moving-average","移动平均线","，就是教科书式的例子。",[161,493,494,497],{},[62,495,496],{},"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。",[161,499,500,503],{},[62,501,502],{},"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。",[161,505,506,509],{},[62,507,508],{},"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。",[14,511,42],{"id":42},[10,513,514],{},"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。",[10,516,517,518,520],{},"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",[47,519,278],{"href":277},"，而且时间够长，经历过开发者从没见过的行情。",[14,522,57],{"id":57},[10,524,525,526,528,529,531,532,528,536,539,540,203],{},"以 ShowMyTrades 上有交易历史的公开账户为样本（2026 年 8 月）：",[47,527,73],{"href":72},"中位数 ",[62,530,77],{},"，历史最深",[47,533,535],{"href":534},"\u002Fglossary\u002Fwhat-is-drawdown","回撤",[62,537,538],{},"9.7%","，每个账户已平仓交易数中位数 ",[62,541,64],{},[10,543,544,545,549],{},"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、",[47,546,548],{"href":547},"\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。",[10,551,552,553,556],{},"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",[47,554,555],{"href":414},"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。",[14,558,101],{"id":100},[10,560,561,562,565,566,569],{},"图表模块里的",[62,563,564],{},"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。",[62,567,568],{},"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。",[10,571,572,573,126,575,122,581,584,585,587,588,126,590,593,594,599],{},"在高级统计模块中，把",[62,574,73],{},[47,576,578],{"href":577},"\u002Fglossary\u002Fwhat-is-standard-deviation-in-trading",[62,579,580],{},"标准差",[62,582,583],{},"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。",[62,586,113],{},"面板里的",[62,589,117],{},[62,591,592],{},"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。",[47,595,597],{"href":596},"\u002Fglossary\u002Fwhat-is-a-verified-track-record",[62,598,390],{},"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。",[14,601,156],{"id":156},[158,603,604,610,616,622],{},[161,605,163,606,609],{},[62,607,608],{},"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。",[161,611,163,612,615],{},[62,613,614],{},"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。",[161,617,163,618,621],{},[62,619,620],{},"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。",[161,623,163,624,627],{},[62,625,626],{},"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。",[10,629,630,631,203],{},"什么样的记录能回答这个问题，什么样的不能，请阅读",[47,632,634],{"href":633},"\u002Fguides\u002Fverified-vs-unverified-track-records","经验证与未经验证的业绩记录",{"title":30,"searchDepth":205,"depth":205,"links":636},[637,638,639,640,641],{"id":16,"depth":205,"text":16},{"id":42,"depth":205,"text":42},{"id":57,"depth":205,"text":57},{"id":100,"depth":205,"text":101},{"id":156,"depth":205,"text":156},"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。",{},"\u002Fzh\u002Fglossary\u002Fwhat-is-curve-fitting",[222,646,448],"what-is-a-profit-factor",{"title":456,"description":642},"zh\u002F5.glossary\u002Fwhat-is-curve-fitting","曲线拟合 (Curve Fitting)","2026-09-10T00:00:00.000Z","seQQ_6nNRjX5XjiWi7epEXN6cZrnk69LgGFRtAA932E",{"id":653,"title":654,"body":655,"category":212,"date":213,"description":821,"draft":215,"extension":216,"meta":822,"navigation":218,"path":823,"related":824,"seo":827,"stem":828,"term":829,"updated":227,"__hash__":830},"glossary\u002Fzh\u002F5.glossary\u002Fwhat-is-trend-following.md","什么是趋势跟踪？定义与真实数据",{"type":7,"value":656,"toc":814},[657,660,662,668,671,677,680,688,690,693,699,701,722,725,731,733,762,778,780,806],[10,658,659],{},"趋势跟踪是一类策略：顺着已经展开的价格走势入场，一直持有到走势结束。它通常赢不到一半的交易，收益靠少数几笔大赚。决定它的是盈亏的形态，而不是入场信号。",[14,661,16],{"id":16},[10,663,664,665,667],{},"入场故意做得很普通：突破 N 日区间、",[47,666,490],{"href":489},"交叉、在排过序的品种池里创新高。真正起作用的是离场：用追踪止损，行情延续就一直拿着，行情停下就回吐固定比例的利润出场。",[10,669,670],{},"之后一切都取决于一个算术等式，平台计算期望值用的也是它：",[21,672,675],{"className":673,"code":674,"language":26},[24],"期望值 = (胜率 × 平均盈利) − (亏损率 × 平均亏损)\n\n趋势示例：  0.35 × 3.0R − 0.65 × 1.0R = 每笔交易 +0.40R\n剥头皮示例：0.85 × 0.4R − 0.15 × 2.0R = 每笔交易 +0.04R\n",[28,676,674],{"__ignoreMap":30},[10,678,679],{},"两个系统都赚钱。第一个差不多每三笔亏两笔，却是更强的那个。只看胜率，不看旁边的平均盈利和平均亏损，这个数字就毫无信息量。",[10,681,682,683,687],{},"收益集中在少数几笔交易里，所以",[47,684,686],{"href":685},"\u002Fglossary\u002Fwhat-is-position-sizing","仓位管理","在这里不是锦上添花，它本身就是策略。在走出大行情的那几笔上仓位太小，负责赚回一切的那段尾部就被砍掉了。",[14,689,42],{"id":42},[10,691,692],{},"这种结构的代价，是大量时间都在亏损。趋势系统会长时间横着走，中间偶尔猛涨一段；人们恰恰是在横着走的阶段放弃它，而且往往就在那波本可以把一切赚回来的行情到来之前。拿一个季度去评判它，看到的只是那个季度的市场，而不是这个系统。",[10,694,695,696,698],{},"这也意味着",[47,697,535],{"href":534},"不是故障。它是入场费，每一次都得提前付。",[14,700,57],{"id":57},[10,702,703,704,69,707,316,709,711,712,714,715,69,717,719,720,203],{},"以 ShowMyTrades 上有交易历史的公开账户为样本（2026 年 8 月），胜率中位数为 ",[62,705,706],{},"68.8%",[47,708,73],{"href":72},[62,710,77],{},"，单笔持仓时间中位数为 ",[62,713,68],{},"。账户已平仓交易数的中位数为 ",[62,716,64],{},[47,718,322],{"href":321},"为 ",[62,721,325],{},[10,723,724],{},"这几乎正好是趋势跟踪的反面：胜率高、持仓短、盈利因子刚过盈亏平衡，典型的靠许多笔小赚去填偶尔一笔大亏。您被拿来比较的这群人实际是什么样子，值得心里有数。",[10,726,727,728,203],{},"样本量也很关键。按 171 笔已平仓交易的中位数算，胜率 35% 的趋势系统大约只有 60 笔盈利单，大部分收益可能就落在其中八到十笔上。拿掉这几笔，记录就是负的。这不是账户脆弱，而是这种盈亏结构注定的结果，也正因如此，这类系统",[47,729,730],{"href":414},"要看几年而不是几个月，数字才有意义",[14,732,101],{"id":100},[10,734,735,736,126,739,122,742,122,745,748,749,754,755,126,758,761],{},"这种结构在高级统计模块里看得最清楚。把",[62,737,738],{},"胜率",[62,740,741],{},"平均盈利",[62,743,744],{},"平均亏损",[62,746,747],{},"期望值","放在一起看：胜率低、平均盈利大，是一个运转正常的趋势系统；胜率高、平均亏损却大得多，则正好相反。",[47,750,752],{"href":751},"\u002Fglossary\u002Fwhat-is-average-trade-length",[62,753,136],{},"能看出持仓周期和宣称的风格是否相符，一个自称“趋势跟踪”、平均每笔却只持仓九分钟的账户，显然是别的东西。",[62,756,757],{},"做多胜率",[62,759,760],{},"做空胜率","能看出这份记录是真正的双向系统，还是押中了某个恰好走出趋势的市场的单边赌注。",[10,763,561,764,766,767,772,773,587,775,777],{},[62,765,535],{},"视图就是水下图：回撤有多深、多频繁，尤其是持续多久，都在上面，而持续时间正是决定一个趋势系统值不值得投资的那个数字。",[47,768,770],{"href":769},"\u002Fguides\u002Fmonthly-returns-table-explained",[62,771,568],{},"会把净值曲线抹平的横盘期和亏损期摆出来；在混合账户上，用",[62,774,113],{},[62,776,121],{}," 筛选，就能把长线持仓和噪音分开。",[14,779,156],{"id":156},[158,781,782,788,794,800],{},[161,783,163,784,787],{},[62,785,786],{},"胜率 35% 说明系统坏了。","”单看这个数字说明不了任何问题。把它和平均盈利、平均亏损放在一起，再看期望值。",[161,789,163,790,793],{},[62,791,792],{},"我把止损收紧一点，胜率就上去了。","”止损越紧，命中率越高，但产生收益的那段尾部也被砍掉了。要把一个赚钱的趋势系统变成亏钱的，这是最可靠的办法。",[161,795,163,796,799],{},[62,797,798],{},"连续八个月表现不佳，说明它失效了。","”对这种结构来说，八个月横盘是正常的间隔，不是判决。",[161,801,163,802,805],{},[62,803,804],{},"胜率 68.8% 的账户更安全。","”高胜率是最容易做出来、也最没有预测力的指标。下任何结论之前，先看平均亏损。",[10,807,808,809,813],{},"上面提到的每个指标，在",[47,810,812],{"href":811},"\u002Fguides\u002Freading-a-trading-account-dashboard","如何读懂交易账户控制台","中都有按模块的逐一讲解。",{"title":30,"searchDepth":205,"depth":205,"links":815},[816,817,818,819,820],{"id":16,"depth":205,"text":16},{"id":42,"depth":205,"text":42},{"id":57,"depth":205,"text":57},{"id":100,"depth":205,"text":101},{"id":156,"depth":205,"text":156},"趋势跟踪就是买强卖弱，用许多笔小亏损换取少数几笔大盈利。为什么盈亏结构比胜率更重要。",{},"\u002Fzh\u002Fglossary\u002Fwhat-is-trend-following",[825,646,826],"what-is-a-moving-average","what-is-drawdown",{"title":654,"description":821},"zh\u002F5.glossary\u002Fwhat-is-trend-following","趋势跟踪 (Trend Following)","ngLD5n2AzFsg76f56aGMBB6kilasnaxpqrdaUrm6vWM",[832,833],{"path":489},{"path":219},1793898068990]