Founder & Philosophy

ABDALLA ELZEDY

310+

RESEARCH SESSIONS
OVER ~2 MONTHS

11,760+

LINES OF
RESEARCH JOURNAL

80+

DETAILED
ANALYSIS FILES

150+

DASHBOARD
DEVELOPMENT SESSIONS

Neural Predictiva was designed, researched, engineered, and deployed by Elzedy. The entire system, from the mathematical foundation through the 12-layer risk pipeline to the real-time monitoring dashboard, represents a coherent vision of what quantitative trading should be.

BACKGROUND

Software engineer and quantitative researcher with deep expertise in applied mathematics, machine learning, and production systems engineering. Background spans full-stack development, data engineering, and algorithmic system design across multiple domains.

Neural Predictiva emerged from a conviction that retail quantitative trading was approaching the problem wrong: too much data mining, too little theory. The research process began with first principles from statistical physics and stochastic calculus, not with indicator screening or pattern matching. Every design decision prioritizes mathematical rigor over backtest performance.

Based in Dallas, Texas. Open to relocating for the right institutional partnership.

THE PHILOSOPHICAL STACK

If you strip away the code, the data, the indicators, and the infrastructure, what remains is a set of beliefs about markets and about how to engage with them.

01

MARKETS ARE NON-STATIONARY

What works now may not work later. Design for adaptation, not permanence.

02

HUMILITY OUTPERFORMS CONVICTION

A system that admits uncertainty, through confidence intervals, probabilistic gating, and regime detection, survives longer than one that does not.

03

DIVERSITY BEATS OPTIMIZATION

Five uncorrelated strategies with moderate individual performance produce better risk-adjusted returns than one heavily optimized strategy.

04

THEORY GENERALIZES; DATA-MINING DOES NOT

Indicators grounded in stochastic calculus, information theory, and state estimation transfer across regimes. Indicators discovered by brute-force pattern matching do not.

05

COSTS ARE REAL

If your edge disappears when you add two pips of round-trip cost, you never had an edge.

06

RISK IS THE ONLY THING YOU CONTROL

You cannot control whether the next trade wins. You can control how much you lose if it does not.

07

SILENCE IS A POSITION

The most profitable thing a trading system can do, on most bars, is nothing.

MARKETS AS WEATHER, NOT CLOCKWORK

Markets are not clockwork. They do not repeat patterns with mechanical precision. But they are also not pure chaos. They exhibit regimes: extended periods where certain statistical properties hold approximately true. The analogy is weather, not clockwork. You cannot predict whether it will rain on March 15th. But you can measure barometric pressure, humidity, and wind patterns right now, and make a reasonable probabilistic statement about the next few hours. That is exactly what this system does with price.

The distinction matters. Prediction implies certainty. Description implies humility. A system built on description adapts when it is wrong. A system built on prediction doubles down.

WHAT THIS SYSTEM IS NOT

NOT A PREDICTION ENGINE

It does not know where any currency pair will be tomorrow.

NOT A BLACK BOX

Every signal can be decomposed into its component indicators, every indicator traced to its mathematical definition, every threshold justified by theory or calibration.

NOT INFALLIBLE

Losses happen. Drawdowns happen. Regime changes happen. The system is designed to survive these, not avoid them.

NOT FINISHED

Markets evolve. Strategies that work today may degrade tomorrow. The system is a living thing, not a monument.

ON BUILDING SOMETHING THAT LASTS

The most difficult part of building a trading system is not the mathematics. It is not the programming. It is not the data engineering or the server configuration. It is the patience to build something honest.

Neural Predictiva is not the most profitable system that could have been built on this data. A system with no regime filters, no gating, no diversity constraints, and no cost model would show a better backtest. It would also fail in production. What was built instead is a system that trades conservatively, admits what it does not know, filters aggressively for quality, and survives.