High-Frequency Analytics

Every trade tells a story.
Most people only see the price.

We analyzed 79 million market messages across 22 trading days. Nanosecond-precision timestamps. 16 message types. Here's what we found hiding in plain sight.

0
Records Analyzed
0
Messages / Day
0
Message Types
0
Query Latency
PTT +1.23% 34.75 ADVANC -0.45% 221.00 SCC +2.10% 368.00 KBANK -1.05% 132.50 SCB +0.78% 98.25 DELTA +3.42% 142.00 AOT -0.33% 62.50 CPALL +0.91% 56.75 TRUE -1.82% 8.95 GULF +1.56% 41.25 BDMS +0.42% 27.00 MINT -0.68% 31.50 PTT +1.23% 34.75 ADVANC -0.45% 221.00 SCC +2.10% 368.00 KBANK -1.05% 132.50 SCB +0.78% 98.25 DELTA +3.42% 142.00 AOT -0.33% 62.50 CPALL +0.91% 56.75 TRUE -1.82% 8.95 GULF +1.56% 41.25 BDMS +0.42% 27.00 MINT -0.68% 31.50
Story 01

The Pump & Dump

A mid-cap stock rises 8% in 25 minutes. The chart looks bullish. But six hidden signals tell a different story.

Price & Volume — 25-Minute Window
Buy-Side Aggressor Ratio
Average Trade Size

"The price said buy. The order book said run."

Manipulation Detected — Confidence: 94% — Flagged at 10:47:32.418
/ Multi-signal scoring — flagged in 7ms
select sym, ret:100*(last price % first price)-1,
  volRatio:sum[quantity]%avg dailyVol,
  buyPct:avg aggressor=`buy,
  avgSize:avg quantity
by sym, 30 xbar time.minute from trade
where date=2025.01.02

Traditional databases take 300ms+ for this query. Our platform returns it in 7ms — fast enough for real-time surveillance across all listed securities simultaneously.

Query Speed — Surveillance Scoring
Our Platform
7ms
Traditional DB
365ms

Six Signals. One Verdict.

Abnormal Return

+8.2% in 25 minutes — exceeds 5% threshold for 30-min window

Volume Surge

4.7x rolling 20-day average volume — concentrated in the pump window

Aggressor Dominance

Buy-side aggressor ratio hits 87% — far above 50% neutral baseline

Shrinking Trade Size

Average trade size drops 62% — classic retail participation pattern

Spread Stability

Bid-ask spread stays tight during run-up — artificial liquidity holding the price

Price Reversal

-6.1% reversal in 8 minutes — dump confirmed as buying pressure collapses

Story 02

The Liquidity Mirage

The order book shows 10 levels deep on both sides. Plenty of liquidity. But decompose the data — and it's the same entity on both sides.

Order Book Depth — 10 Bid/Ask Levels
Order Book Imbalance Over Time

"10 levels of depth. Zero real liquidity. The order book was a stage set — and the same actor played both buyer and seller."

/ Wash trade detection — matching bid/ask patterns
select count i, sum quantity
by sym, 5 xbar time.minute
from trade where date=2025.01.15,
  bidQty within (0.95*askQty;1.05*askQty)
Story 03

The Halt Aftermath

A trading halt lifts. In the first 200 milliseconds, the order book tells you everything about what happens next. If you can read it fast enough.

Price & Spread — 5 Minutes Around Halt

What happens in 200ms?

  • >The equilibrium price shifts 2.3% from pre-halt reference
  • >Bid-ask spread widens 4x before snapping back
  • >First 50 trades are 90% sell-aggressor
  • >Order book depth rebuilds in 1.2 seconds

"You can't see this in a daily chart. You can't see it in minute bars. You need nanosecond timestamps and a database that can query them in real time."

Story 04

The Arbitrage Window

Equity index vs. futures. A basis spread opens for 800 milliseconds. Only tick-level cross-market data reveals it.

SET Index vs. Futures Basis Spread — Intraday

"By the time a human sees the opportunity, it's gone. The data saw it 800ms earlier."

/ Cross-market basis spread — as-of join
aj[`time;
  select time, spot:price from indexPrice;
  select time, fut:price from futTrade]

As-of joins across markets in sub-millisecond time. This is what columnar in-memory engines are built for.

Story 05

The VWAP Hunter

An institutional order is slicing through the market all day. The trades look random. But overlay VWAP and the algo's fingerprint appears.

Price vs. VWAP — Full Trading Day
Trade Size Distribution
/ VWAP — computed in 7ms across 10.4M trades
select vwap:wavg[quantity;price]
by sym, 1 xbar time.minute
from trade where date=2025.01.02
The Problem

Traditional databases weren't built for this.

3.6 million messages per trading day. 16 message types. Nanosecond timestamps. Order book reconstruction. Multi-signal aggregation in real-time.

Row-based databases choke on ingestion. Generic analytics engines can't do as-of joins. Embedded databases run out of memory. Each tool has trade-offs.

So we tested them all.

The Method

We Tested 4 Database Architectures

Same data. Same queries. Same machine. No marketing claims — just measured performance on real market data.

A — Columnar In-Memory B — Columnar On-Disk C — Embedded Analytical D — Traditional RDBMS
Default Configuration
After Optimization

Default Configuration

Query A B C D
OHLCV 1-min27ms65ms82ms125ms
Range scan57ms104ms123ms198ms
Market-wide agg61ms70ms52ms147ms
Top-N by volume56ms69ms52ms122ms
Surveillance scoring90ms82ms101ms364ms
VWAP calculation22ms67ms57ms83ms

After Optimization

Query A B C D
OHLCV 1-min7ms65ms33ms122ms
Range scan21ms104ms47ms196ms
Market-wide agg47ms70ms23ms145ms
Top-N by volume38ms69ms23ms137ms
Surveillance scoring65ms82ms34ms365ms
VWAP calculation7ms67ms22ms76ms

Storage Footprint

DatabaseSizeNotes
B0.17 GBBest compression
C1.24 GBColumnar, embedded
D1.33 GBRow-based
A6.03 GBIn-memory, fastest queries

Ingestion Speed (10.4M trades)

DatabaseTimeRate
B43s243k rows/s
A145s72k rows/s
D243s43k rows/s

"There is no single best database. A dominates time-series queries. C wins on analytical aggregation. B leads on compression and ingestion. D is familiar but consistently slowest. The right answer depends on your use case — and we help you make that decision."

Independent benchmark for educational purposes. 85 million records, 22 trading days, p50 latency over 5 runs, all databases containerized on the same machine. Results reflect our test environment and may vary.

The Platform

Built for Nanosecond-Precision Analytics

The engine behind these stories was chosen after rigorous benchmarking. Here's why it works.

ns

Nanosecond Timestamps

Native nanosecond precision — matches exchange protocol timestamps exactly

||

Columnar Storage

Column-oriented architecture ideal for time-series aggregations across billions of rows

In-Memory + On-Disk

Hot data in RAM for real-time, historical data on disk — seamless hybrid architecture

q>

Expressive Query Language

Purpose-built for financial analytics — as-of joins, window functions, vector operations in one line

90%

Proven at Scale

Used by the majority of global tier-1 banks and exchanges for mission-critical workloads

Py

Python Integration

Seamless Python bridge for ML pipelines, data science workflows, and dashboard integration

Architecture

End-to-End Pipeline

From raw exchange feed to actionable insight — fully containerized, deployable with one command.

Exchange Feed

16 message types
Nanosecond precision

Parser

Protocol decoder
Normalization

Analytics Engine

Columnar in-memory
Partitioned by date

Analytics Layer

Surveillance
Market analytics

Dashboard

Interactive UI
Real-time alerts

Columnar In-Memory DB Python Docker Parquet REST API Interactive Dashboard
Applications

Any Industry. Any Event Stream.

Capital markets is just the demo. The same platform analyzes any high-frequency event data — wherever milliseconds matter.

Capital Markets

  • >Real-time market surveillance across all listed securities
  • >Manipulation detection — pump & dump, spoofing, wash trading
  • >Best execution & transaction cost analysis
  • >Regulatory compliance reporting on tick-level data

Energy & Utilities

  • >Millions of smart meter readings per hour
  • >Grid sensor streams at sub-second intervals
  • >Load anomaly detection & demand spike forecasting
  • >Outage cascade identification before propagation

Telecommunications

  • >Billions of CDRs and packet events per day
  • >Real-time DDoS pattern detection
  • >Service degradation identification & root cause
  • >Capacity allocation optimization

Manufacturing & IoT

  • >Thousands of sensors streaming at microsecond intervals
  • >Predictive maintenance — detect failure before it happens
  • >Multi-sensor signal correlation for root cause analysis
  • >Real-time quality control on production lines

Transportation & Logistics

  • >GPS ticks from thousands of vehicles, every second
  • >Real-time route optimization & ETA prediction
  • >Delivery anomaly & geofence violation detection
  • >Driver behavior pattern analysis & fleet utilization

Healthcare & Life Sciences

  • >ICU waveform data at millisecond intervals
  • >Real-time cardiac anomaly & sepsis onset detection
  • >Clinical trial event stream analytics
  • >Pharma manufacturing quality control

The pattern is always the same: massive volume, nanosecond precision, real-time decisions.

Why Us

We Don't Just Sell Tools. We Choose Them.

Platform-agnostic means we benchmark objectively, recommend honestly, and deliver the right solution — not the most expensive one.

Since 2016

End-to-end Data, AI, and Automation consultancy — not a reseller pretending to consult

🏳

Thai Market Expertise

Deep understanding of local exchange data structures, trading rules, and regulatory requirements

Platform-Agnostic

We tested 4 database architectures on real data. We recommend based on evidence, not partnerships

Full-Stack Delivery

From data pipeline to analytics engine to production dashboard — one team, end to end

AI

AI Integration

Layer ML, anomaly detection, and natural language queries on top of your analytics platform

Deploy Anywhere

Fully containerized — cloud, on-premise, or hybrid. One command to deploy

Get Started

What stories are hiding
in your data?

Let us show you what's underneath. Live demo on real high-frequency data — or bring your own.

Request a Demo

🌏 We welcome international engagements — serving clients across Southeast Asia and beyond.

contact@infozense.com  |  +66-82-242-4008  |  Bangkok, Thailand