PreTrAIde Methodology: How a Pre-Trade Analysis Is Computed

What the analysis evaluates

A user builds a basket of one or more stocks, specifying a buy or sell side, quantity and execution conditions for each line. PreTrAIde analyzes the proposed basket before the user places any order with their own broker. One analysis uses one analysis credit.

The methodology has three layers:

• Data enrichment supplies market data, company information, recent news and analyst ratings.
• A deterministic quantitative layer computes trading-cost and risk estimates, portfolio breakdowns and 0–100 scores. Given the same inputs, its calculations are reproducible.
• An AI layer interprets the stock-level and basket-level evidence in plain language. Guardrails check its output against specified constraints and can make a verdict more conservative.

These layers answer different questions. A spread estimate concerns the likely cost of crossing a quoted market; volatility describes price variability; concentration describes how basket weights are distributed. A verdict brings those considerations together but does not replace them or establish that a trade will succeed.

Data sources and enrichment

PreTrAIde enriches each stock with about 57 fields drawn from market data—including end-of-day prices, volumes and company profile information—plus recent news for the ticker and analyst ratings. The quantitative layer uses the market-data inputs to calculate estimates and scores. The AI layer can also consider news and ratings when explaining the stock and basket assessments.

Coverage includes US stocks and major European exchanges, such as Euronext Paris, Xetra and the London Stock Exchange. Data availability and quality can differ across stocks. PreTrAIde computes a data-quality score so that limitations in the inputs can be considered alongside the resulting estimates. A news item or rating is context, not proof that a price will move in a particular direction.

Estimated bid-ask spread

The bid-ask spread is the difference between the best displayed offer and bid at a given moment. It is one component of execution cost: a buyer typically pays the offer, while a seller typically receives the bid. PreTrAIde estimates spread using a liquidity-tier model, a volatility uplift and a one-tick floor. This is an estimate for pre-trade analysis, not a live quote or a guaranteed fill.

The tier is based on average daily dollar volume (ADDV): average shares traded per day multiplied by share price, expressed in US$ per day. Higher ADDV generally indicates more trading activity, but it does not guarantee that a particular order can be filled at a given price. The baseline spread assumptions are:

A basis point (bp) is 0.01%; 100 bp is 1%. Before the tick floor, the baseline is multiplied by a volatility uplift:

Volatility uplift = min(1 + daily volatility in percentage points / 2, 3)

For example, if hypothetical daily volatility is 2%, the uplift is min(1 + 2/2, 3) = 2. A 5 bp baseline becomes a 10 bp estimate before the tick floor. The multiplier is capped at 3. When applying the formula, 2% is entered as 2, not 0.02; mixing percentage points and decimals would materially change the result.

A tick is the minimum permitted price increment for a stock on its market. The one-tick floor prevents an estimated spread from falling below that increment. Because a tick is denominated in price units, its equivalent in basis points depends on the stock price. For illustration, a hypothetical US$0.01 tick at a US$20 share price equals 5 bp of the share price.

Kyle (1985) and Glosten-Milgrom (1985) are foundational market-microstructure references for understanding how trading, information and liquidity relate to prices and spreads. They provide context for the estimate; they should not be read as a claim that a tier-based estimate reproduces the displayed spread at any specific time.

Participation and estimated market impact

Participation measures an order's size relative to typical daily trading. A simple expression is:

Participation rate = proposed shares / average daily shares traded

If a hypothetical order is 10,000 shares and average daily volume is 1,000,000 shares, participation is 10,000 / 1,000,000 = 1%. This comparison does not specify how quickly the order will be executed. A 1% share of an entire day's volume could be much harder to trade within a few minutes than across a longer period.

Market impact is the potential price effect associated with executing an order. PreTrAIde estimates it with an Almgren-Chriss-style model. In models with square-root behavior, impact grows with order size but less than proportionally over a given range: doubling a size-dependent term increases it by approximately √2, or 1.41, rather than exactly 2. This describes model behavior, not a promise about a real fill.

Spread and impact are related but distinct. Spread concerns the cost of transacting across bid and offer; impact concerns how trading itself may affect the attainable price. Execution timing, available depth, other participants' orders and changing market conditions can make realized costs differ from both estimates. Almgren-Chriss (2000) is a reference for the trade-off between execution cost and risk over time; the label “Almgren-Chriss-style” does not mean that every real-world execution follows its assumptions.

Daily volatility: four estimation perspectives

Daily volatility describes the scale of daily price variation, usually expressed as a percentage of price. PreTrAIde blends four estimator types rather than relying on a single measurement:

• Close-to-close volatility uses changes between successive closing prices. It is straightforward, but a close-only series does not separately describe what happened within each session.
• Parkinson volatility uses daily high and low prices to capture intraday range information. Its usefulness depends on the assumptions behind a range-based estimator and the reliability of the high and low data.
• Yang-Zhang volatility draws on open, high, low and close information and is designed to account for overnight and intraday variation.
• A simplified GARCH(1,1) estimator models volatility as varying over time, with recent shocks and prior estimated volatility contributing to the current estimate. “Simplified” matters: it is a model-based perspective, not an observation of future volatility.

Blending helps avoid treating any one view of historical price movement as complete. It does not remove model risk. A sudden event, a trading halt or a change in market conditions can make past-data-based volatility a poor guide to the next session. Higher estimated volatility also feeds the spread uplift described above.

One-day 95% value at risk

PreTrAIde computes one-day 95% value at risk (VaR). In its volatility-based expression, the estimated VaR rate is:

1-day 95% VaR rate = 1.645 × daily volatility

If daily volatility is a hypothetical 2% of position value, the VaR rate is 1.645 × 2% = 3.29%. For a hypothetical US$10,000 position, that corresponds to about US$10,000 × 0.0329 = US$329 of estimated one-day downside at the 95% threshold, under the model's assumptions.

95% does not mean the maximum loss is US$329. In this example, outcomes beyond the threshold remain possible, and losses beyond it could be substantially larger. The calculation is sensitive to the volatility estimate and to assumptions implicit in using a normal-distribution multiplier. It does not account perfectly for abrupt gaps, unusual market conditions or every form of tail risk. VaR is a risk measure, not a forecast of what a stock will do tomorrow.

Liquidity tiers, scores and data quality

The ADDV table assigns a baseline spread tier. PreTrAIde also computes liquidity scoring and tiering as part of its broader quantitative assessment. A tier groups stocks for estimation; a score expresses a model assessment on a 0–100 scale. Neither is a live measurement of order-book depth, a guarantee of tradability or a substitute for checking the actual market when placing an order.

Data quality matters alongside liquidity. An apparently precise output can still depend on incomplete or unrepresentative inputs. The data-quality score indicates the quality of the available data used in the analysis; it does not certify that all relevant information has been captured.

Basket composition and concentration

PreTrAIde breaks down a basket by country, sector, region, currency and market capitalization. These views show where proposed positions are grouped. They do not, by themselves, measure every economic relationship between the companies.

The Herfindahl-Hirschman Index (HHI) measures concentration using position weights:

HHI = sum of squared basket weights

With weights written as decimals that total 1, HHI ranges from 1/N for an equally weighted basket of N positions to 1 for a basket entirely in one position. A hypothetical basket weighted 50%, 30% and 20% has HHI = 0.50² + 0.30² + 0.20² = 0.38. If percentage-point weights are used instead—50, 30 and 20—the equivalent HHI is 3,800 on a 0–10,000 scale. The two conventions describe the same concentration; they must not be mixed.

HHI captures position-size concentration, not all forms of common exposure. Two equally weighted stocks in the same sector can produce a relatively low position-weight HHI while still sharing sector-specific risks. That is why the portfolio breakdowns and the index answer complementary questions.

Quantitative scores and AI reasoning

PreTrAIde computes 0–100 liquidity, volatility and execution scores for each stock and for the basket. They make the quantitative assessment easier to compare across lines and at basket level. A score is a model output, not a probability of profit: a 70 execution score does not mean a 70% chance of a favorable return.

AI agents then reason about each stock and the whole basket, anchored on the quant scores, recent news and analyst ratings. Anchoring matters because a persuasive written explanation should not override the calculated evidence without being checked. The quantitative layer is deterministic and reproducible; the AI layer supplies contextual interpretation and plain-language reasoning.

The output includes a GO, NO-GO or MODIFY verdict, a risk assessment, a suggested execution strategy and a success probability. That probability is a model estimate bounded between 5% and 95%. It is neither a guarantee of an execution outcome nor a forecast of returns. In particular, it should not be read as the chance that a stock's price will rise or that a basket will make money.

Guardrails on the verdict

PreTrAIde applies guardrails when AI output contradicts specified data or rules. A suspended stock is always NO-GO. A GO with a success probability below 50% is downgraded. Probabilities are clamped to the 5%–95% range.

Guardrails can move a verdict only in the conservative direction; they do not upgrade a verdict to make it more favorable. These rules address identified contradictions, not every possible data error or market event. GO, NO-GO and MODIFY remain decision-support labels, not instructions to trade or to refrain from trading.

Limits, uncertainty and independence

Pre-trade analysis is necessarily an estimate made before execution. PreTrAIde uses end-of-day prices and volumes among its market-data inputs; those historical observations may not represent the spread, depth, volatility or news environment at the moment an order is placed. Estimated spread and impact can differ from actual execution costs, particularly when conditions change or an order is large relative to available liquidity.

Each model also simplifies reality. The volatility blend depends on historical prices and model assumptions. The normal multiplier in VaR may understate unusual outcomes. ADDV measures typical activity rather than executable liquidity at a specific time. Basket concentration measures selected dimensions of exposure, not every source of shared risk. News and analyst ratings may be incomplete, stale or open to interpretation. Model estimates and plain-language reasoning should be read with these uncertainties in mind.

PreTrAIde is analysis software, not a broker, investment adviser or portfolio manager. It does not receive, transmit or execute orders and has no order flow from users' trades. Users decide what to do with the analysis and place any trades themselves through their own broker. The methodology is intended to make the inputs, estimates and limits of a pre-trade decision clearer—not to promise a result.