几乎所有延迟套利账户的情况都是如此。最初几周看起来很不错：订单执行干净利落，交易持续几秒钟，而杠杆率曲线则呈稳定上升趋势。然而，在没有任何电子邮件或警告的情况下，情况发生了变化。订单执行开始出现延迟，延迟幅度达到半个点，而且总是针对你的不利因素。曾经只需 40 毫秒即可执行的订单，现在需要 300 毫秒才能执行。账户上的点差看起来比经纪商公布的差值略高一些。几个星期后，监管部门发出通知称正在进行“交易活动审核”。.
这一切都不是运气不好。这是一个检测系统在执行其工作。本指南将对此进行解释。 经纪人如何检测延迟套利 从交易者的角度来看：风险团队实际计算的是哪些数字，为什么在这些数字中套利模式显得格外突出，账户被标记后通常会出现怎样的升级过程，以及如何阅读自己的交易记录，以判断它是否真的发生在你身上。.
如果您对策略本身还不太熟悉，请先从我们的视频开始学习。 关于延迟套利原理的解释. 本文假定您已了解基本原理：将慢速报价与更快的参考数据源进行交易。.
为什么经纪人一开始就会寻找延迟套利机会？
侦查是跟资金有关的。一个用流动性提供方对每一笔交易进行对冲的经纪商（纯粹的 A-book 模式），对你的策略几乎不感兴趣；但其 LP 可能会有所不同，因为 LP 最终会承担掉这些价格滞涨带来的损失。而一个承担客户交易另一方（B-book 或混合模式）的经纪商，则会将每笔套利获利视为自己的损失。一个快速账户就可以抹平来自数百名普通客户的差价收入。.
所以，风险团队提出的问题并不是真的“这是否是一种套利行为？”而是“这种流动对我的投资有毒吗？”毒性高的流动是指交易的结果，这些结果能持续预测价格的走向。延迟套利是这种现象的最清晰的例子，这也是为什么它如此容易衡量的原因。关于这一过程的路由环节，请参阅我们的详细分析。 A-book与B-book经纪人.
最重要的指标：标记
如果经纪公司的风险团队只能查看一个数字，那它就会是 标记在每次平仓后的短暂时间内，市场会向您有利的方向大幅波动。经纪商会在您平仓时采用中价，随后再在 100 ms、500 ms、1 秒和 5 秒后再次采用中价，并将所有交易的差价平均计算出来。.
对于普通零售客户来说，平均手数接近于零，或者包含在手数范围之外的交易。零售交易者平均来说无法预测价格行为的下一秒。延迟套利账户的情况则截然不同。其交易行为完全不受延迟的影响。 因为 价格即将变动，因此标记曲线几乎同时跳跃起来，然后平缓下来。这种形状很难通过偶然方式产生。.
Markout之所以强大，是因为它无视了你的故事。它并不在乎你是否把你的系统称作是套利者、新闻交易者还是格栅系统。它只关注你的成交量是否系统性地领先于价格，而正是这种优势延迟套利机制所依赖的。.
其他信号风险引擎相结合
Markout 单独也会产生误报。经验丰富的新闻交易员有时也会生成正面标记。因此，现代风险管理工具会将其与一系列行为特征相结合。这些工具中的任何一种都无法单独证明套利行为的存在。它们共同作用下，可以得出非常可靠的得分。.
| 信号 | 它测量什么 | 为何套利模式脱颖而出 | 你自己的日志中会看到什么 |
|---|---|---|---|
| Markout | 填充后的 0.1-5 秒内价格出现上涨 | 交易之所以开启，是因为这一举动已经开始进行。 | 大多数交易会在一秒内就获利。 |
| 持有时间分布 | 头寸保持开放的时间有多长 | 秒数紧凑，而不是分钟到天的分布情况 | 单次握持时间平均为数以秒计 |
| 按持有时间计算的胜率 | 特别是在非常短期的交易中获胜率 | 短线交易通常是将点差减去，剩下的就是获利；大多数情况下，套利策略都能获利。 | 高获胜率集中在最快的交易中 |
| 订单的执行时间与点差 | 参考市场价格更新与您的订单之间的延迟时间 | 订单会在外汇市场交易结束后不久即到达，在此之前经纪公司的报价才能够跟上。 | 帖子集中在快速移动和新闻热点上 |
| 失速不对称 | 正差与负差的比值 | 一旦账户被标记为违规，经纪商就可以控制这一比例，因此这也是一个结果。 | Positive slippage disappears, negative slippage grows |
| Sizing and pattern regularity | Repeated lot sizes, symbols, session times | Bots repeat themselves; humans drift | Same lot on nearly every trade, same few symbols |
| Cross-account links | Shared IPs, devices, mirrored or opposite P&L | Two-leg and multi-account set-ups leave correlated footprints | Not visible to you, but it is checked |
From rules to models
Early detection was a set of hard rules: “flag any account with an average hold time under N seconds and a win rate over X%.” Rules like that are easy to step around and they catch plenty of innocent scalpers. Most serious brokers and their technology vendors now feed the features above into a scoring model that ranks accounts by toxicity, often refreshed daily or even intraday. The practical effect is that no single threshold exists to stay under. The model weighs the whole profile, and the profile of a pure latency strategy is distinctive.
What happens after an account is flagged
Brokers rarely close an account the moment it is flagged. Closing is the loudest option and it ends the spread income too. What usually happens is a quiet escalation, each step designed to take the edge away while the account keeps trading. Knowing the ladder matters, because each rung leaves a visible trace in your results.
Step 2: execution changes
The account is moved to a different execution group. Orders may be held for a short delay before they are filled. A delay of even a hundred milliseconds is enough to erase an edge that lives in the first few hundred. Slippage becomes one-way: when the price moves against you during the delay you get the worse price, and when it moves for you the order is filled at the requested price or rejected. This is what a “virtual dealer” set-up does, and it is the most common response by far.
Step 3: pricing changes
Next the account may get a wider spread markup than the published one, a slower or smoothed price feed, or a last-look window in which the broker or its LP can reject fills that have become unprofitable for them. On paper your strategy still “works”; in practice the expected value per trade falls to around zero.
Step 4: account action
Finally comes the visible step: a trading review, reduced leverage, a block on certain instruments, a closed account, or, in the worst case, cancelled profits under an “abusive trading” or “price latency exploitation” clause in the client agreement. Read that clause before you deposit, because it decides whether a dispute is even worth having.
How to tell if your account has already been flagged
Brokers don’t announce steps 1 to 3, but you can measure them. You have the same data they do for your own trades. Run this check every week:
- Compare execution time over time. Export order-sent and order-filled timestamps. If your median fill time jumps from tens of milliseconds to hundreds, and the jump is not explained by your VPS or network, the account was most likely moved.
- Count positive versus negative slippage. On a fair account both happen. If positive slippage drops to almost nothing while negative slippage grows, the execution is asymmetric.
- Track profit per trade, not just win rate. A falling average profit with an unchanged signal count means the edge is being taxed somewhere between your order and the fill.
- Watch the reject and requote rate. A sudden rise in “off quotes” or “price changed” errors on fast moves is the classic last-look footprint.
- Compare the account’s spread to the published spread. Log the bid/ask at entry. A persistent extra fraction of a pip is a markup, not market conditions.
Any one of these can be noise. Two or three together, starting in the same week, are a strong sign that the account has climbed the ladder. The rational move then is usually not to fight the broker but to withdraw and redeploy.
What this means for anyone running arbitrage
Detection is not a reason to give up on latency arbitrage. It is a reason to run it as a professional operation and not a lottery ticket. In practice that means:
- Pick the venue first. The detection model matters less when the broker’s business model does not suffer from your profit. Our guide on 如何选择套利经纪商 covers what to check: execution model, LP set-up and the wording of the client agreement.
- Read the terms, literally. If the agreement explicitly bans exploiting price latency, you are on borrowed time there, however good the first month looks.
- Withdraw often. Profits sitting in an account are the profits that can be reviewed. Regular withdrawals limit what a late-stage review can touch.
- Keep your own logs. Timestamps, prices and slippage per trade are your evidence in any dispute, and they are also your early-warning system.
- Expect a lifecycle. Plan for an account to have a useful life measured in weeks or months, and size your effort accordingly.
Traders who want to understand the counter-measures that strategy developers have built in response to these models (trade-pattern variation, multi-account exposure rotation and similar techniques) will find them covered in our article on 新一代套利交易技术. Whatever approach you take, the broker’s own terms remain the rulebook that applies to your account.
常见问题解答
如果经纪人能够检测到延迟差价，那么是否属于非法行为？
Detection and legality are different questions. In most jurisdictions trading on a broker’s slow quote is not a criminal matter. It is a contractual one, governed by the broker’s client agreement. We cover the legal side in detail in is forex arbitrage legal?
经纪公司需要多长时间才能发现延迟套利行为？
It depends on trade volume and how sophisticated the broker’s tooling is. Markout and hold-time statistics need a sample of trades to become reliable, so an active account at a broker with modern risk software can be scored within days. At a smaller broker that reviews accounts manually it can take weeks.
如果经纪人发现存在套利行为，能否取消我的利润？
Some client agreements give the broker the right to cancel trades or profits judged to come from price latency or “abusive” trading. Whether that holds up depends on the exact wording and on the regulator. That is why reading the agreement before depositing, and withdrawing regularly, matter so much.
仅仅使用快速的VPS或低延迟就能让我的账号被封禁吗？
No. Low latency by itself is not a signal. Brokers want fast clients who trade normally. What gets an account flagged is what the speed is used for: consistently positive markout, very short hold times and high win rates on those short trades.
经纪人是否会分享关于套利交易者的信息？
Brokers that use the same white-label technology, bridge or liquidity provider can end up with similar views of toxic flow, and LPs see flow from many brokers. Don’t assume a new account at a related broker starts from a clean slate.
所有出现点差账户都是被标记的账户吗？
No. Slippage is normal in fast markets. The warning sign is a sudden change in the pattern: slippage becoming almost entirely negative, together with slower fills and more rejects, starting at roughly the same time.
作者简介
