AI trading vs algorithmic trading
AI trading uses models (including reinforcement learning) to learn decision rules from data; algorithmic trading can be purely rule-based. Here’s the practical difference.
Algorithmic trading: rules you write
Algorithmic trading is the broad category: you define a strategy and automate execution.
An algorithmic strategy can be as simple as “buy when price crosses above a moving average, sell when it crosses below”. It can also be complex (multi-signal, portfolio constraints, risk overlays). The key point: the decision logic is explicitly designed and implemented by you.
AI trading: rules a model learns
AI trading usually means using ML models to learn patterns and decisions from data.
In AI trading, you typically specify inputs (features), an objective, and constraints, and then train a model to produce decisions (signals, actions, allocations). Reinforcement learning trading is one slice of this: an agent learns sequential decisions by interacting with an environment and optimizing a reward function.
What changes in practice
- Evaluation matters more — it’s easy to overfit. Walk-forward testing and strict leakage controls are not optional.
- Data quality is a first-class dependency — garbage in, garbage out, amplified.
- Monitoring is part of the strategy — once live, you need drift detection and run-time safety checks.
Where Kabu fits
Kabu is built for reinforcement learning trading workflows.
You define the environment, actions, observations, and reward; train agents in backtests; then deploy a fixed policy to live markets with monitoring. The goal is to keep research and production connected: experiments, artifacts, and live execution in one place.