Security Selection

​

The Selection Pipeline

Security Selection runs as a cascade. The eligible universe enters the pipeline, and each step scores the securities it receives, applies its selection rule, and passes the survivors to the next step. A security must pass every step to be selected. This cascading design makes each score relative to its context. A security that looks above-average on value among all 2,400 names in the universe may look below-average among the 200 quality survivors of an earlier step. Because each step re-scores within the survivors it receives, the score always reflects a security’s standing among the names still in contention.

​

Long and Short Legs

A strategy selects securities on one or two legs:

• Long Position: the long leg always runs. Its final survivors are the securities the model holds long.
• Short Position: the short leg runs only for long-short strategies. It is configured independently, with its own steps, metrics, and rules.

​

Steps and Cascading

Each leg is a sequence of up to three steps. Steps run in order, and each step narrows a single survivor set that feeds the next. Every step has two parts:

1. Scoring: which metrics define the score, and how they combine into one number.
2. A selection rule: how that score is turned into a keep-or-drop decision.

Use multiple steps when the selection logic is genuinely sequential; when a later ranking should only be measured among the survivors of an earlier screen. Use a single step when one combined score already expresses the full intent.

​

Security Scoring

​

Selection Rules

The selection rule converts the combined score into the step’s survivor set. There are two rule types, matching the two filter types.

​

Relative Ranking

A Relative Ranking selects securities based on their position in the ranked list of combined scores. At each reconstitution date, BIASafe takes the securities entering the step, calculates their combined scores, ranks them, and keeps the part of the ranking defined by the user. This can be a fixed number of securities, a percentage of the universe, or a band within the ranking. The cutoff is relative to the score distribution at that step. It is not an absolute score value. For example, selecting the Top 20% does not mean the score must be above a fixed number. It means BIASafe keeps the best-ranked 20% of securities available at that reconstitution date, regardless of where the actual score cutoff falls.

​

Ranking Method

The Ranking Method defines whether the filter keeps a fixed number of securities or a fixed proportion of the ranked universe.

• Sequential Rank: keep a fixed count (e.g. the top 100).
• Percentile Rank: keep a fixed percentage (e.g. the top 20%).

​

Direction

The Direction defines whether higher or lower scores are preferred.

• Higher Is Better: larger scores rank first. This is commonly used for metrics where higher values are preferred, such as momentum, profitability, growth, or quality scores.
• Lower Is Better: smaller scores rank first. This is commonly used for metrics where lower values are preferred, such as volatility, valuation multiples, leverage, drawdown, or risk measures.

How Direction and Slice Work Together

​

Slice

The Slice defines which part of the ranking is kept after scores are ordered.

• Top: keep the best end of the ranking.
• Bottom: keep the worst end of the ranking.
• Between: keep a middle band defined by a lower and an upper cutoff.

​

Threshold Filtering

A Threshold Filter selects securities based on whether their combined score passes a defined cutoff. Unlike a Rank Filter, a Threshold Filter does not keep a fixed number or percentage of securities. Instead, it keeps every security whose score satisfies the selected comparison rule. The final number of selected securities can therefore change at each reconstitution date, depending on the score distribution. For example, a strategy may keep every security with a combined score greater than 0, or every security above the median score of the eligible universe.

​

Comparison Type

The Comparison Type defines how the cutoff is determined.

• Fixed Threshold: compares each security’s combined score against a value entered by the user. For example, if the rule is Greater Than 0, BIASafe keeps every security whose combined score is positive.
• Dynamic Reference: compares each security’s combined score against a statistic calculated from a reference population. For example, the rule may keep every security with a score above the mean or median score of the eligible universe.

Dynamic Reference Settings

​

Reference Scope

For Dynamic Reference, the Reference Scope defines the population used to calculate the threshold.

• Eligible Universe: Uses the securities entering the current selection step. This means the dynamic threshold is calculated only from the securities that survived the previous filters and steps.
• Starting Universe: Uses the original universe at the beginning of the selection leg, recalculated using the current step’s metrics. This allows the threshold to be anchored to the broader starting universe instead of only the securities that reached the current step.

​

Reference Statistic

For Dynamic Reference, the Reference Statistic defines which statistic is used as the cutoff.

• Mean: Uses the average combined score of the reference population.
• Median: Uses the middle combined score of the reference population.

The median can be useful when the score distribution is skewed or affected by outliers.

​

Comparison Rule

The Comparison Rule defines how each score is evaluated against the selected threshold. Available rules include:

• Greater Than: Keeps securities with scores above the threshold.
• Greater Than or Equal: Keeps securities with scores at or above the threshold.
• Less Than: Keeps securities with scores below the threshold.
• Less Than or Equal: Keeps securities with scores at or below the threshold.

​

Turnover Buffer

The Turnover Buffer helps reduce unnecessary changes at each reconstitution. Without a buffer, a security can be removed simply because it falls slightly below the selection cutoff. This can create turnover even when the difference in score is marginal. With a buffer, new securities must still clear the normal entry cutoff, but existing holdings are given more room before they are removed. This creates a buffer zone that helps retain current holdings unless their ranking deteriorates more meaningfully.

​

Buffer Zone

The turnover buffer works by combining the model’s existing selection cutoff with an additional buffer threshold. The buffer threshold defines how far an existing holding can drift before it is removed. The area between the selection cutoff and the buffer threshold is the buffer zone. It prevents the model from replacing an existing holding with a new security when the difference in rank is marginal. Because of this, new securities and existing holdings are treated differently:

​

Buffer Type

The Buffer Type defines how the buffer width is measured. It can be expressed two ways: Use a Rank Buffer when the strategy targets a fixed number of securities. Use a Percentile Buffer when the strategy targets a percentage of the eligible universe. This allows the buffer to scale as the universe grows or shrinks.

​

Replacement Method

The Replacement Method defines how BIASafe handles the final portfolio count when existing holdings are retained inside the buffer zone.

Replacement Method Settings

​

Keep Incumbents (default)

It retains all existing holdings that remain inside the buffer zone, even if this causes the portfolio to hold more securities than the normalThis option provides the strongest turnover reduction, but the number of holdings may vary within the buffer range.For example, if the model normally selects the top 100 securities, the portfolio may temporarily hold more than 100 securities when several existing holdings remain inside the buffer zone.

​

Replace All Incumbents

Replace All keeps the portfolio count fixed at the selection cutoff.Existing holdings inside the buffer zone are still given priority, but they replace the weakest new entrants one-for-one. This preserves the turnover-control benefit of the buffer while keeping the number of holdings fixed.Use this option when the strategy requires an exact number of securities at each reconstitution.

​

Tie Handling

Tie Handling defines how BIASafe resolves securities that have the same score at the boundary of a Rank Filter. A Rank Filter keeps a fixed-size slice of securities, such as the top 100, top 20%, or a middle-ranked band. Because scores are not always unique, multiple securities may share the exact same score at the cutoff point. When this happens, BIASafe needs a rule to decide which securities remain inside the selected slice.

Tie Handling Settings

​

Include All Ties

Include All Ties keeps the full group of securities tied at the boundary. This is the most neutral option when securities are indistinguishable based on the selected score. The final selection may be larger than the target count.For example, a top 100 filter may return 103 securities if four securities are tied for the final available position.Use this option when it is preferable to include a few additional equivalent securities rather than force an arbitrary cutoff.

​

Random Tie-Break

Random Tie-Break fills the remaining seats by randomly selecting from the tied securities. By default, the selected securities may differ across runs when ties occur. Users may provide a random seed to make the result reproducible. With the same seed, the same tied securities are selected each time.Use this option only when random selection is acceptable for the strategy design.

​

Deterministic

Deterministic resolves ties using a secondary tie-break metric. Users select:

• Tie-Break Metric: The metric used to order tied securities, such as Market Cap, Average Daily Dollar Volume, Beta, or Volatility.
• Preference: Whether the highest or lowest value of the tie-break metric receives priority.

If securities remain tied after the secondary metric is applied, BIASafe uses the security identifier as a final stable tiebreaker. This ensures that the same configuration produces the same selection every time.

​

Missing Value Policy

A metric may be unavailable for some securities on a given selection date. The Missing Value Policy defines how BIASafe treats those securities before ranking and selection are applied. This setting is important because missing data can affect the eligible universe, the ranking order, and the final securities selected by the model. In most cases, Exclude from Ranking is the safest default because it avoids making assumptions about securities with unavailable data. Other policies may be used when the strategy has a clear reason to keep securities in the selection step despite missing values.

​

Machine-Learning Overlays