FollowSM Engineering

Standard Z-Score relies on mean and standard deviation. In crypto, a single liquidation cascade creates extreme outliers that skew the mean and inflate the standard deviation. This causes standard Z-Score to miss subsequent toxic volume spikes.
Robust Z-Score replaces mean and standard deviation with Median and MAD (Median Absolute Deviation), making it immune to outliers:
$$\text{Robust Z} = 0.6745 \times \frac{x - \text{Median}}{\text{MAD}}$$
1. The Math Problem: Standard Z-Score vs. Outliers
Assume a rolling window of orderbook trade volumes: [10, 12, 11, 15, 100, 14] (where 100 is a sudden liquidation spike).
Standard Z-Score for a new volume of 25:Mean = 27.0, Std Dev = 32.8Standard Z = (25 - 27.0) / 32.8 = -0.06 (Fails to trigger an alert!)Robust Z-Score for a new volume of 25:Median = 13.0, MAD = 2.5Robust Z = 0.6745 × (25 - 13.0) / 2.5 = 3.24 (Triggers a toxic volume alert!)
2. Pure Python Implementation (Standard Z-Score)
If you calculate volume anomalies manually without MAD, your execution baseline breaks during high volatility:
def calculate_z_score( list[float]) -> float: if not data or len(data) < 2: return 0.0 mean = sum(data) / len(data) variance = sum((x - mean) ** 2 for x in data) / len(data) std = variance**0.5 return (data[-1] - mean) / std if std else 0.0 # Example usage: volumes = [10.0, 12.0, 11.0, 15.0, 100.0, 25.0] z = calculate_z_score(volumes) print(f"Standard Z-Score: {z:.2f}") # Output: -0.06 (Misleading)
3. Production-Ready SDK Alternative
Skip manual orderbook aggregation and outlier math using @followsm/sdk:
npm i @followsm/sdk
import { FollowSMClient } from "@followsm/sdk"; // Initialize client (omit apiKey for free tier) const client = new FollowSMClient({ apiKey: "fsm_live_..." }); // Fetch real-time orderbook toxicity & Robust Z metrics const snapshot = await client.getToxicitySnapshot("BTCUSDT"); if (snapshot.is_toxic_alert) { console.log(`⚠️ Toxic Flow! VPIN: ${snapshot.vpin}, Volume Robust Z: ${snapshot.volume_z_score}`); }
4. Real-time Snapshot Payload
{ "symbol": "BTCUSDT", "timestamp": 1758412800.0, "price": 62150.5, "vpin": 0.72, "ob_toxicity_1pct": 2.35, "ob_imbalance_l1": 0.61, "depth_bands": { "0.5%": { "bid_notional": 184320.0, "ask_notional": 96410.0, "imbalance_ratio": 0.657 }, "1.0%": { "bid_notional": 312500.0, "ask_notional": 210800.0, "imbalance_ratio": 0.597 }, "2.0%": { "bid_notional": 590100.0, "ask_notional": 470200.0, "imbalance_ratio": 0.556 } }, "volume_z_score": 3.23, "natr_15m": 0.84, "taker_buy_ratio": 0.58, "is_toxic_alert": true }
#Crypto #TradingBot #Quant #BinanceSquareTalks #HFT $BTC $ETH

🛡️ Beyond Candle Charts: How Market Makers Spot Orderbook Toxicity (VPIN) Before Flash Crashes
Ever wondered why order books suddenly thin out seconds before a massive crypto dump? It’s not magic — it’s order flow toxicity.
If you're building trading bots, market-making strategies, or execution algorithms, relying solely on standard time-based candles (RSI, MACD) leaves you blind to millisecond-level asymmetric information.
Here is how institutional market makers detect toxic order flow using VPIN — and how you can apply it.
📉 Why 1-Minute Candles Deceive You
Time-based intervals aggregate trades linearly. But crypto liquidity is not linear:
During quiet hours, 1,000 trades might take 30 minutes.During a liquidation cascade, 1,000 trades happen in 50 milliseconds.
When informed traders or institutional algorithms enter the market, market makers face adverse selection — they get filled on the wrong side of the trade right before price shifts aggressively.
📊 What is VPIN (Volume-Synchronized Probability of Toxicity)?
To catch toxic flows, quantitative trading uses Volume Buckets instead of time intervals.
Volume Synchronization: A new bucket closes only after a fixed volume V is traded (e.g., every 100,000 USDT).Order Imbalance: Trade flow inside each bucket is split into Buy (V_buy) and Sell (V_sell) volume.Toxicity Index: VPIN measures the rolling average of imbalance across N buckets:
VPIN = Σ | Buy_Volume - Sell_Volume | / (N × Bucket_Volume)
When VPIN spikes above historical baselines (e.g., > 0.70), it signals extreme toxic flow — informed buyers/sellers are aggressively sweeping liquidity.
💡 How Quants Use VPIN to Protect Capital
Dynamic Spread Skewing: Spreads are widened immediately when VPIN jumps, avoiding toxic fills.Execution Pause: TWAP / VWAP execution algorithms hold order placement during toxicity spikes to cut slippage.Automated Risk Breakers: HFT systems trigger automatic inventory reduction.
🛠️ Hands-on Microstructure Metrics
If you're writing custom strategy engines in Python or Node.js/TypeScript, you don't need to rebuild orderbook bucket processing from scratch.
You can pull low-latency orderbook toxicity & smart money metrics via standard SDKs:
Python (PyPI):
pip install followsm-sdk
from followsm import FollowSMClient client = FollowSMClient() # Fetch real-time toxicity metrics for BTCUSDT metrics = client.get_toxicity_snapshot("BTCUSDT") if metrics.vpin > 0.70: print(f"⚠️ Toxic flow detected! VPIN: {metrics.vpin:.2f}. Pausing execution.")
#Crypto #TradingBots #Quant #BinanceSquare #MarketMicrostructure #BTC $BTC $ETH$SOL