01
AI Quant Model
Deep learning extracts signals from market and fundamental data; pricing and timing are decided by the model.
TIANMA QUANT · HONG KONG
Let the model read the market
Let discipline become instinct
Tianma Quant applies deep learning to fully automated trading across global equity and fund markets. Every decision, from signal to execution, is traceable and verifiable.
SCROLL
02
Fully Automated Trading
From signal generation and risk constraints to order execution — no human intervention, every decision traceable.
03
AI Research
Sustained investment in compute, operators and training frameworks, so iteration is never limited by resources.
THE MODEL
A system iterated for
thirteen years
In 2013 the first quant model began trading proprietary capital. In 2016 a deep-learning model went live. By 2019 every strategy had been converted to AI.
In 2023 the system moved beyond A-shares, using the Hong Kong Hang Seng market as its primary back-testing ground. V2 and V3 followed in 2024; the milestone release of May 2025 benchmarks alongside leading closed-source models at a fraction of their training cost.
2013
Origin
ORIGIN
12yr
Iteration
ITERATION
3gen
T+0 models
GENERATIONS
1sys
Automated system
SYSTEM
PERFORMANCE
Compounding comes
from discipline
Annual returns stepped up as the architecture matured: 51% in 2019, 77% in 2021, 125% in 2025.
Figures are based on internal company statistics. Past performance is not indicative of future results and does not constitute investment advice.
HISTORY
Thirteen years,
one thread
From a small team exploring fully automated trading to a system that runs across markets. Every step answered a concrete constraint.
RESEARCH
Compute serves
research
The ceiling of a model is set by training and back-testing throughput. To break the single-machine bottleneck of 2017, the team built its own GPU cluster — so an idea can be tested the same week it is proposed.
The market does not become predictable because models get smarter. But a model can make every bet closer to mathematics.