udt
updowntrends
How-To Guide

Understand AI stock forecasts & trading signals — from metrics to execution

Learn how to evaluate forecast quality, identify market regimes, and interpret signals across 1D, 5D, 20D, and 60D — with transparent validation per asset.

All analyses and forecasts are generated automatically using AI and technical analysis. They are for informational purposes only and do not constitute investment advice.

Metrics Overview

Metrics Overview – evaluate forecast quality

The Metrics Overview is the entry point to objectively evaluate AI forecast quality. Select the base (e.g., all, stocks, or a specific index), the prediction horizon, and the top selection. This helps you quickly see where the model historically produced robust signals.

The horizon determines whether you analyze short-term signals (1–5 trading days) or more stable mid-term horizons (20–60 trading days). AVG is a weighted average across horizons (1D 0.20 · 5D 0.25 · 20D 0.25 · 60D 0.30).

Interpretation: Use metrics as a quality filter. Higher values indicate better separation. Lower values suggest being more cautious or selecting a different horizon.

Definitions are explained in the Technology FAQ.

Metrics Overview
Performance Overview

Performance Overview – understand market regime

This view shows historical performance of the selected assets over a defined period (e.g., 1Y to 1D). It provides the context needed to interpret forecasts properly.

Tooltips provide per-asset details. The pie chart summarizes whether market breadth was mostly positive or negative — useful to distinguish trending vs drawdown phases.

Interpretation: In clear trend regimes, mid horizons (20–60) are often more stable. In volatile regimes, switching horizon can help reduce noise.

Performance Overview
Predictions Overview

Predictions Overview – market barometer per horizon

This overview shows all predictions for the selected assets on the chosen horizon. It gives you a fast market snapshot: how many assets are UP, DOWN, or neutral?

Select the horizon (1 to 60 trading days) to separate short-term moves from mid-term trends. Tooltips show per-asset signal and UP probability.

Interpretation: Many aligned signals on the same horizon indicate broad strength. Strongly mixed signals often point to sideways or transition phases.

Predictions Overview
Prediction Table

Prediction Table – sort, search, compare

The prediction table provides a detailed overview across all assets. Every column is sortable (e.g., accuracy, performance, direction). Search takes you directly to the asset detail page.

Interpretation: A proven workflow is: sort by quality first, then filter by direction, and validate on the asset detail page.

Prediction Table
Asset Page

Asset Performance – trend, regime & strength

The performance chart shows an asset’s historical development. You can identify trend direction, reversals, and move strength — the most important context before interpreting predictions.

Interpretation: When trend and prediction align, signals are often more robust historically. Counter-trend signals can work, but tend to be more regime- and volatility-dependent.

Choose the horizon that matches your time frame to avoid confusing short-term noise with mid-term trends.

Asset Performance
Asset Page

Asset Predictions – probabilities & transparency

The predictions table shows forecasts for horizons from 1 to 60 trading days (plus AVG). For each horizon you see the signal (UP/DOWN/UNCLEAR) and the UP probability — turning forecasts into probabilistic statements rather than rigid yes/no calls.

Transparency: Horizon-specific quality metrics are shown. This lets you evaluate not only the signal, but also historical reliability for this asset and horizon.

Note on 0% or NA: If a metric shows 0% or NA, it could not be computed reliably in the selected window — typically because there isn’t enough history (e.g., at least 60 trading days for 60D metrics).

For definitions & methodology, see the Technology FAQ.

Asset Predictions
Asset Page

Prediction History – stability over time

Prediction history shows UP probabilities and signals over time. Select a horizon to see whether the view was consistent or changed frequently.

Interpretation: Stable clusters indicate more reliable signals. Isolated spikes are often short-term market noise.

Prediction History
Asset Page

Prediction Quality: Signals & Hits – hits per horizon

This chart shows generated signals and hits per horizon. It helps you quickly identify which horizons were historically more reliable for this asset.

Interpretation: A horizon with consistently more hits is often the better choice — even if other horizons show stronger but less reliable spikes.

Signals & Hits
Asset Page

Prediction Quality: Correlation – prediction vs performance

The correlation plot compares normalized prediction values (p_raw − 50%) with normalized performance values (lag). Visually, it shows whether predictions and later price moves historically aligned.

Interpretation: A visible positive structure indicates genuine forecast quality. A diffuse cloud suggests weak or regime-dependent signals — consider switching horizon or revisiting the asset selection.

Correlation
Metrics Development

Metrics Development – quality over time

This view shows how a selected quality metric evolves over time. It helps detect stability, improvements, or regime shifts that can influence forecast quality.

For methodology and definitions, see the Technology FAQ.

Metrics Development
Articles

Understand signals & apply them step by step

These guides cover the questions investors search for most: what the UP probability means, how signals are formed, how to combine quality and context — and how to translate that into better entry, exit, and risk-management decisions.

1) How are signals generated — and how should I interpret them?

The core idea is simple: updowntrends estimates the probability that a price will rise over a defined forecast horizon. The UP probability is the key value because it shows not only direction but also how strong the statistical conviction is. A corresponding DOWN probability can be derived — the higher UP is, the lower DOWN becomes.

Important: the computation is based on historical end-of-day (EOD) data and derived features (e.g., indicators, volume signals, market structure). This tends to make signals more stable than pure intraday impulses. At the same time, it’s not a ‘100% future prediction’ — it’s a probability statement that behaves like a data-driven tailwind/headwind indicator in practice.

Why do signals require clarity thresholds?

To keep signals usable, the system intentionally reduces noise. An UP signal typically appears only when clarity is high (e.g. ≥ 70% UP). A DOWN signal appears when the UP probability is very low (e.g. ≤ 30%). In between, the situation is often statistically ambiguous — and that’s useful: you avoid constant pseudo-signals when the data has no clear edge.

How do you read UP and DOWN in practice?

An UP signal means: in historically similar situations, the price moved up more often than down over the horizon. A DOWN signal means: it moved down more often than up. The main value is not the arrow — it’s the probability. It helps you weight decisions instead of making them binary.

What is this especially good for?

Signals are especially valuable in two common situations: (1) timing for fundamentally selected stocks: you know *what* you want — signals help with *when*. (2) risk management: you can see whether your time horizon has statistical tailwind or whether you should plan more defensively (position sizing, scaling, patience).

Next: the following article shows how to combine signals with quality (metrics) and context (trend/regime) to make systematic entry & exit decisions.

2) Using signals in detail: validate quality, select candidates, decide entry & exit

The best way to use AI signals is with a clear workflow. You don’t start with ‘UP or DOWN’ — you start with: ‘How well did this work historically — for this asset and this horizon?’ That’s what metrics are for. They show whether probabilities were meaningful in the past and whether the model separated UP and DOWN phases.

Step-by-step workflow

Step 1: quality before direction. Sort by quality first (robust metrics) and identify horizons that work cleanly for an asset. Step 2: context before action. Check trend and performance: trend regime, sideways regime, or a volatile transition? Step 3: signal as decision support. Only then does it make sense to use direction and probability — not as a trigger, but as weighting.

Entry: timing without overtrading

A sensible entry is rarely ‘buy immediately because UP’. Often it’s smarter to use signals as a timing layer: if 20D/60D show tailwind and the trend is stable, pullbacks can be cleaner entry zones than chasing extended moves. The multi-horizon view helps you see whether short-term momentum (e.g. 5D) supports the structure (20D/60D).

Exit: react earlier without exiting too early

Signals are especially helpful for spotting risk early. Two patterns are practical: (1) probability flips meaningfully *and* the history shows frequent signal changes — often a sign of messy regimes. (2) longer horizons turn even though short-term strength remains — often a hint that risk–reward is deteriorating. This can justify reducing position size, taking partial profits, or re-evaluating the setup.

Using limitations correctly (instead of fearing them)

The key point is: signals work best in stable regimes and when quality metrics confirm the model historically performs well for this asset and horizon. That’s why transparency (metrics, signal history, hits) and multi-horizon views matter: you don’t just get ‘UP’ — you get an interpretable picture that supports better decisions.

3) Combining AI forecasts: fundamentals, charts and timing

One of the most common search questions is: ‘Does AI replace traditional analysis?’ The strongest use case is the opposite: AI complements fundamentals and charting. You use fundamentals to pick quality — and AI forecasts to improve timing: when entries are statistically more attractive, when patience is better, and when risks are rising.

A practical three-layer framework

Layer 1: fundamentals (quality, growth, valuation) — answers ‘what belongs in the portfolio?’. Layer 2: chart structure (trend, support/resistance, volume) — answers ‘where does it get interesting?’. Layer 3: AI probability per horizon — answers ‘when is risk–reward statistically more favourable?’

When these layers align, you get setups that are far calmer to execute. You avoid chasing, you can interpret pullbacks as potential entries within a trend, and you often get earlier signals when the setup deteriorates for your time horizon. That’s what the platform is built for: less gut feeling, more decision structure.

Another benefit is speed: instead of manually screening dozens of charts, you can prioritise ideas data-driven — and spend your time where quality, context, and probability converge.

4) Working with multiple horizons: focus on structure, not noise

Many investors don’t struggle because of missing information, but because of poor weighting. This is where multiple forecast horizons are a real advantage. A 5-day signal answers a different question than a 60-day signal — both can be valuable, but for different decisions.

Short-term vs mid-term vs long-term

Short-term horizons react quickly to momentum and sentiment. Mid horizons (e.g. 20D) help identify whether a move is more than a brief impulse. Longer horizons (e.g. 60D) are especially useful for recognising regimes and steering strategic exposure — whether to build, hold, or become more defensive.

A typical scenario: the market corrects short-term while 20D and 60D stay positive. Multi-horizon signals help avoid treating pullbacks as trend breaks. Conversely, turning long-term horizons can provide early hints that risk is increasing — even if short-term strength remains visible.

The biggest benefit is separating tactical from strategic: short horizons help fine-tune, longer horizons steer position sizing, patience, and risk. This leads to more consistent decisions — which is exactly what many investors search for when looking up ‘AI trading signals’ or ‘AI stock prediction strategy’.

5) Risk management with AI signals: position sizing, stops, drawdowns

Many people search ‘how to use AI signals safely’. The best answer isn’t a perfect entry — it’s clean risk management. Signals help you manage risk more deliberately, especially via position sizing, scaling, and choosing the right horizon.

Position size instead of all-in

If 20D/60D are clearly UP and quality metrics are strong, you can often take risk more efficiently. If horizons conflict or history flips frequently, the better decision is often not ‘none’ but ‘smaller’. Position sizing is the lever that stabilises outcomes — whether you trade short-term or invest long-term.

Planning stops & exits sensibly

Signals don’t replace stop logic — they improve it. A common mistake is setting stops too tight when the regime is volatile. If history shows strong swings and signals flip faster, plan more defensively: smaller size, more room, clear partial-profit or re-entry plan. If the trend is stable and longer horizons show tailwind, stops can often be structured around trend levels more confidently.

Bottom line: AI signals are strongest when they improve discipline. They help you avoid jumping into every move, dose risk better, and reduce drawdowns — and that’s the core of many successful strategies.

6) Earnings, news & macro: what signals can do — and how to plan event risk

A very common search topic is ‘Do AI signals work before earnings?’ The right framing is: signals estimate probabilities based on historical patterns. Big events (earnings, central bank decisions, geopolitical news) can dominate the market short-term — which is why event risk should be a separate planning step.

How to still use signals effectively

When an event is coming up, use signals mainly for context: does the setup fit the broader trend? Is there tailwind on 20D/60D? How stable is signal history? This doesn’t yield ‘certain up’ or ‘certain down’ — it helps you plan risk: smaller size, clear exit rules, or deliberately waiting until the event passes.

This is where probabilistic signals shine: you don’t have to guess. You can categorise decisions: ‘high clarity + stable trend’, ‘mixed + event-driven’, ‘unclear + volatile’. This tends to produce better outcomes than purely reactive trading.

Definitions for probabilities, metrics, hit rate, walk-forward and out-of-sample are explained in the Technology FAQ.