From AI Signals to Decisions: Build Risk Rules (Not Predictions)

In practice, AI is rarely the hard part. The hard part is decision-making under uncertainty.

Many people treat model outputs as “answers.” But a prediction without constraints is just a number that invites impulse. A usable workflow turns outputs into rules—explicit limits that prevent a single bad assumption from dominating your portfolio.

Here’s a simple, implementation-friendly framework you can adapt:

Step 1 — Specify the risk budget
Before you consume any signal, define your allowable downside (portfolio max drawdown, per-position loss cap, or volatility cap). If the risk budget is undefined, the model can’t help you, because you haven’t defined what “safe” means.

Step 2 — Convert signal strength into position size
Instead of “buy/sell,” treat signals as sizing suggestions. Stronger confidence can scale exposure up, but only within strict bounds.

Step 3 — Add a rebalance rule
Rebalancing is a “boring algorithm” that often beats emotional discretion. Use time-based or threshold-based triggers.

Below is a toy example to illustrate the mindset (educational only):

def size_position(signal, max_weight=0.10):
# signal expected in [-1, 1]
base = 0.02 # 2% minimum meaningful size
weight = base + abs(signal) * (max_weight - base)
return min(weight, max_weight)

def rebalance(current_weight, target_weight, band=0.02):
return abs(current_weight - target_weight) > band

Step 4 — Log decisions for feedback
A model improves when your process produces clean feedback. Record: signal, size, risk rule, and outcome. Over time, you’ll learn whether the model is useful—or whether your rules need refinement.
https://www.rimbamindaai.com/