Methodology

Design principles, assumptions, limitations.

Siddhantha is an analytical platform, not a forecasting service. This page sets out the principles behind how the platform ranks stocks, what it assumes, and what it cannot do.

Design principles

The T20 platform uses proprietary machine-learning models trained on point-in-time market data. For each universe and each horizon, the platform evaluates both the upside and downside case for every stock, producing a ranked list of high-conviction long and short ideas. We deliberately separate long and short evaluation so that each is scored on its own merits rather than being reduced to a single number.

Signal construction

Raw market data is transformed into a proprietary library of signals covering price behaviour, trading activity, relative strength within the universe, and relationships to broader indices and sector benchmarks. Signals are computed point-in-time — we never use information that would not have been available at the moment the pick is made.

Refresh cadence

The analytical engine is refreshed on a rolling basis so it continuously adapts to changing market conditions. Picks are regenerated daily (T20 Daily) or weekly (T20 Weekly) before they are published to the platform.

Ranking

The platform's primary ranking metric is a conviction indicator, which captures how strongly the analytical engine favours one direction over the other for a given stock. High conviction indicates a clear analytical signal. Low conviction indicates ambiguity, which we treat as a signal to stay neutral.

Assumptions

Assumption	Implication
Markets are partially analysable	The platform assumes cross-sectional return patterns persist long enough to be identified and acted on.
Historical conditions inform current conditions	If the market regime changes abruptly and without precedent, performance will degrade.
Liquidity is sufficient at top-of-book	Picks assume users can transact in the named tickers without large slippage.
No transaction costs in raw scores	Rankings are pre-cost. Users must account for commissions, spread and impact.

Limitations

The platform cannot anticipate idiosyncratic news (earnings surprises, M&A, regulatory action) until that information appears in market behaviour. The platform is not a market-timing tool — it ranks relative attractiveness within a universe, it does not call the direction of the broader market. Past performance is not a reliable indicator of future results.

Continuous improvement

We continue to research new signals, new analytical approaches, and better evaluation techniques. When a material change reaches production, we publish release notes so users can understand what has changed.

Learn

Probabilities, models, and what they can and cannot tell you.

Siddhantha is a probabilistic system. Used well, that is a strength. Used badly, it leads to the wrong conclusions. This page sets out the core concepts.

Probabilities, not predictions

A model that outputs "stock X has an 81% probability of outperforming tomorrow" is not promising that stock X will outperform tomorrow. It is saying that, over many days where the model assigned the same probability, roughly 81% of those stocks did in fact outperform. The probability is a long-run statement, not a single-event guarantee.

This means a single losing day from a high-confidence pick is not evidence that the model is wrong. It is evidence consistent with the 19% of the time that high-confidence picks underperform.

Why model assumptions matter

Every model is trained on historical data and assumes that the relationships in that history will continue to hold. When markets undergo a regime change — a new monetary regime, a new dominant factor — the model may take time to adapt. During that adaptation period, performance can degrade.

What model-based analytics is good for

Cross-sectional ranking — picking the relatively most attractive names within a defined universe — is what the Siddhantha models are designed for. They are not designed to call the direction of the broader market, time entry and exit at the index level, or replace fundamental due diligence on individual companies.

How to use the platform well

Treat the T20 as a starting universe, not a buy list. Combine it with your own fundamental review, your own position sizing, and your own risk management. The model's job is to surface candidates worth your attention — what you do with them is your judgment.

About

A UK-based quantitative platform built around guiding principles.

Siddhantha LTD is a UK-registered company building probabilistic stock-intelligence tools for serious investors. We do not manage client money. We publish model output. The user decides what to do with it.

What we believe

Markets are partially predictable. Cross-sectional patterns persist long enough that machine-learned signals can offer meaningful edge — provided they are honestly calibrated, transparently presented, and consumed by users who understand their limits. That is the discipline we hold ourselves to.

Why "Siddhantha"

"Siddhantha" is a Sanskrit word meaning established conclusion or guiding principle. The platform is built around guiding principles — first to protect principal, then to capture asymmetric opportunity. The brand reflects the discipline.

Who we are

Siddhantha is built by a small team of quantitative researchers and engineers based in the UK. Backgrounds span machine learning, equity research and high-frequency trading infrastructure.

Disclaimer

Siddhantha publishes the output of statistical models. This material is for informational purposes only and does not constitute investment advice, an offer or solicitation to buy or sell any security, or a representation that any investment strategy is suitable for any particular individual. All investment carries risk, including loss of principal. Past performance is not a reliable indicator of future results. See the legal page for full terms.