Over the last month, I tested 100+ arbitrage/news bots for Predict Markets using the best arb strategies, and the result shocked me…
**| Before we start, bookmark this and drop a follow
| Posting daily alpha on Polymarket and more
There are 2 types of the most effective assistants/bots for trading on prediction markets
1. News bot **
2. Arbitrage bots
News bots usually have the same system but use different data sources (YouTube, Twitter, Reddit, news pages on-chain whale analytics)
Real event: news drops (T = 0s)
News Bot: parses source, analyzes sentiment, places trade (T=+0.5-2s) Polymarket odds: start moving (T = +3-7s)
Human trader: notices the news, opens position (T = +15-30s)
Δt ≈ 13-28s opportunity window
Why these bots are effective and viable - because there's a ton of these opportunities in the market tied to some news
per month 1000+ opportunities ( min per day 30+)
This gives a huge room for research and bot testing. Usually, they all use these simple math formulas
1. Mispricing Function ( M(p,t) = C(p,t) - p/100 )
2. Empirical Kelly ( f = f_kelly × (1 - CV_edge) )
3. VPIN Toxicity ** ( |Vb - Vs| / (Vb + Vs) )*
Arbitrage bots run fully automated 24/7 non-stop parsing Polymarket Kalshi API
Real event: ETH dump -4.2% on Chainlink Data (T = 0s)
FastLoop (Arb Bot): detects difference between Polymarket and Kalshi, places trade (T = +0.3s)
Polymarket odds: start moving (T = +2-3s)
Human trader: notices the difference between platforms, opens position (T = +8s)
Δt ≈ 7.7s - opportunity window
usually using the simplest and most effective formulas
- Arbitrage Invariant ( P(YES) + P(NO) = $1.00 )
- Mispricing Function ( M(p,t) = C(p,t) − p/100 )
- Frank-Wolfe Profit ( max_δ[min_u(δ·φ(u) − C(θ+δ) + C(θ))] )
- Empirical Kelly ( f = f_kelly × (1 − CV_edge) )*
Bots are easy to implement, you just need an API and a good server that runs well
downside is the low number of trades per month 80+
Yesterday I launched one, and in 2 hours, it found only 4 trades +1-3%
I ran detailed analytics on which markets have more arbitrage opportunities
Over the last year:
- Politics 1200+
- Sports 800+
- Crypto 670+
- Economy 670+
Almost all of them use ClaudeCode as the central brain for decision-making, after finding a trade Claude decides whether to open it or not
delegate trade finding and position sizing to an assistant bot - it'll easily handle that using EV and Kelly
I personally prefer a system like telegram notifications for finding a trade where you make the decision yourself
The optimization era has arrived - either you adapt, or you just won't make it in time to open the position and get your dreamed life change
Link: http://x.com/i/article/2026578225875255296
过去一个月里，我用最好的套利策略测试了 100+ 个用于预测市场的套利/新闻机器人，结果让我大吃一惊……
| 在开始之前，把这篇文章收藏并关注我
| 每天分享 Polymarket 等的 alpha
预测市场交易中最有效的两类助手/机器人只有两种：
1. 新闻机器人
2. 套利机器人
新闻机器人 通常系统架构相同，但会使用不同的数据源（YouTube、Twitter、Reddit、新闻网站、链上巨鲸分析等）
真实事件：新闻发布（T = 0s）
新闻机器人：解析来源、分析情绪、下单（T = +0.5-2s）
Polymarket odds：开始变动（T = +3-7s）
人类交易者：注意到新闻并开仓（T = +15-30s）
Δt ≈ 13-28s 机会窗口
为什么这些机器人有效且可行——因为市场里与新闻绑定的机会实在太多了
每月 1000+ 次机会（每天至少 30+）
这给研究与机器人测试提供了巨大的空间。通常它们都会用这些简单的数学公式：
1. 定价偏差函数（ M(p,t) = C(p,t) - p/100 ）
2. 经验凯利（ f = f_kelly × (1 - CV_edge) ）
3. VPIN 毒性（ |Vb - Vs| / (Vb + Vs) ）*
套利机器人 则是全自动 24/7 不间断运行，持续解析 Polymarket / Kalshi API
真实事件：ETH 在 Chainlink Data 上暴跌 -4.2%（T = 0s）
FastLoop（套利机器人）：检测到 Polymarket 与 Kalshi 的差异并下单（T = +0.3s）
Polymarket odds：开始变动（T = +2-3s）
人类交易者：注意到平台差异并开仓（T = +8s）
Δt ≈ 7.7s - 机会窗口
通常使用最简单、也最有效的公式：
- 套利不变量（ P(YES) + P(NO) = $1.00 ）
- 定价偏差函数（ M(p,t) = C(p,t) − p/100 ）
- Frank-Wolfe 利润（ max_δ[min_u(δ·φ(u) − C(θ+δ) + C(θ))] ）
- 经验凯利（ f = f_kelly × (1 − CV_edge) ）*
机器人很容易实现：你只需要一个 API，以及一台运行稳定的好服务器
缺点是每月交易数量偏少：80+ 笔
昨天我上线了一个，2 小时里只找到 4 笔交易，收益 +1-3%
我做了详细分析：哪些市场有更多套利机会
过去一年：
- 政治 1200+
- 体育 800+
- 加密 670+
- 经济 670+
几乎所有系统都会用 ClaudeCode 作为决策中枢：发现交易后，由 Claude 决定是否开仓
把找交易和仓位管理交给助手机器人——它用 EV 和 Kelly 就能轻松处理
我个人更喜欢用 Telegram 通知来发现交易机会，由你自己做决策的系统
优化时代已经到来——要么适应，要么你就来不及开仓，更别提实现你梦寐以求的生活改变
链接: http://x.com/i/article/2026578225875255296
