Portfolio Rebalancer

An important premise. QANTHOS is an analysis, validation and discovery tool, not a financial advisor. The thresholds, criteria and parameters described in this chapter are configurable levers: they establish how the Rebalancer intervenes, not how much concentration is “right” for you. That judgement depends on your capital, your instruments and your risk tolerance: you are always the final judge.

1. Concentration: a silent source of risk

It’s worth dwelling on why this module exists, because the problem it solves is easy to underestimate. A portfolio of systematic strategies performs best when its sources of return are independent: if each strategy reacts to different dynamics, its drops don’t coincide with the others’, and overall equity ends up smoother than that of the individual components.

Concentration works against this principle. If many strategies pile onto the same ticker — or the same asset class — they’re no longer independent sources: they share the same fate. A single adverse event on that instrument hits them all together, and the portfolio amplifies the blow instead of absorbing it. Drawdown gets deeper, volatility higher, and what looked like diversification (many strategies) turns out to be an illusion (one single bet, repeated).

The Portfolio Rebalancer is the safeguard against this illusion. It doesn’t judge strategy quality — that’s the filters’ job — but their distribution: it makes sure risk stays spread across an adequate number of different groups, instead of clumping onto a few.

2. The logic: dominance over the average of the others

This is the module’s central design choice, and it’s the same philosophy behind the Percentile Rank Filter: reasoning in relative terms, not absolute.

One could imagine a Rebalancer that works with a fixed cap: “no more than 2 strategies per ticker, period”. Simple, but rigid. That 2 knows nothing about the rest of the portfolio: in a small, already concentrated portfolio two strategies on the same instrument can be a lot, while in a large, well-distributed portfolio they’re perfectly normal. A number carved once and for all doesn’t distinguish the two cases — and ends up being too strict in one and too permissive in the other.

QANTHOS’s Rebalancer instead adopts an adaptive rule, dominance over the average of the others. For each group (a ticker, or an asset class) it counts how many strategies populate it and compares it with the average count of the other groups at the same level. A group is dominant when its count exceeds that average multiplied by a threshold you set:

count(group)  >  threshold  ×  average( count of OTHER groups )

The threshold is a multiplier, not a cap. With threshold 2.0 you’re saying: “the most represented ticker can reach at most double the average of the other tickers; beyond that, it’s badly disproportionate and needs lightening”. With threshold 1.5 you’re stricter (one and a half times the average is enough), with 3.0 more tolerant. There’s no maximum number of strategies decided upfront: the bar moves on its own, following the shape of the portfolio.

An example clarifies the difference (numbers are illustrative, depend on your pool). Imagine that, after filters, the other tickers bring an average of two strategies each. With threshold 2.0 a ticker becomes dominant as soon as it accumulates more than four (two for the average, two for the multiplier). If instead the other tickers’ average rose to three, the same threshold would allow up to six strategies on the same ticker before intervening. The rule stays the same, but the effective limit breathes with the context — and it’s precisely this adaptivity that makes the tool valuable.

Why relative beats absolute, here. A fixed cap protects against one scenario and ignores another. Relative dominance, instead, asks a question that always makes sense regardless of portfolio size: is this group really out of scale compared to its peers?. It’s the same reason the Percentile Rank Filter prefers “the top 30% of the current group” to “Sortino above 0.8”: a self-calibrating threshold keeps working even when absolute conditions change.

When a group is dominant, the Rebalancer removes its least deserving strategy — according to the exclusion criteria you configure — then recalculates everything and starts again. It’s an iterative and adaptive process: each exclusion changes the counts, which changes the average of the others, which changes the effective threshold for the next round. It stops when no group exceeds its own threshold anymore. Sizing (Money Management) happens afterwards, on the surviving strategies, so size redistributes across an already-balanced portfolio.

When the top groups are two (or more), tied. There’s a case the “average of the others” rule, alone, would handle badly. Imagine two groups that, after filters, end up with the exact same number of strategies — the two most populated in the portfolio, perfectly matched. If judging the first included the second in the average, and vice versa, the two would shield each other: each raises the other’s average, and the comparison makes both look normal, even when together they weigh disproportionately on the portfolio. To avoid this, when two or more groups are tied at the top, QANTHOS treats them for what they are — the leaders — and excludes them from the average of the others: the average is computed only on the smaller groups, the ones that genuinely represent the rest of the portfolio. If the leaders exceed the threshold computed this way, they’re lightened together, one strategy per group at each pass with a recalculation after each, so they stay balanced with each other as they come back into line.

An example (illustrative numbers). At the asset-class level, two families bring eight strategies each and a third brings four, with threshold 1.5x. The average of the smaller groups is four, so the limit is 1.5 × 4 = 6. The two leading families, both at eight, exceed it: the Rebalancer removes one strategy from each and recalculates, then again, stepping down in lockstep (8 and 8 → 7 and 7 → 6 and 6). Once at six neither exceeds the limit anymore and the process stops — with a much more balanced portfolio than would have resulted from treating the two leaders as each other’s shield (in that case both would have stayed at eight).

The weight of the rest: the “other groups” quorum. Pushing the leaders toward the limit only makes sense if there’s a rest substantial enough to serve as a reference. If below the leaders there remained just one tiny group of a single strategy, using it as the yardstick to reduce all the others would do more harm than good: most of the portfolio would be dismantled chasing a proportion dictated by a detail. That’s why the Rebalancer only intervenes when the “other” groups (those below the leaders) are numerous enough to really count: by default they must represent at least one third of the total groups. Below this threshold the level is left as is. It’s a configurable parameter (see Activation Conditions): raising it demands a bigger rest before acting, lowering it lets the Rebalancer intervene even with few reference groups.

An example of the brake in action (illustrative numbers). With four groups of 8 / 8 / 8 / 1, the only “other” group is the one with one strategy: it weighs a quarter of the total, below the one-third threshold, and the Rebalancer doesn’t intervene. Changing the composition to 8 / 8 / 1 / 2, the “other” groups become two out of four — half the total, above the threshold — and rebalancing resumes. The same mechanism also acts as a natural brake: as leaders get lightened and the composition flattens, the “other” groups thin out, the quorum is no longer met, and the Rebalancer stops on its own, without pushing the portfolio toward forced uniformity. And when all present groups have the same number of strategies — an already-perfect balance — there’s nothing to correct: the Rebalancer recognizes the condition and stops.

3. Where it acts in the chain

The Rebalancer occupies a precise position in the Portfolio Rules chain: it intervenes after the filters and after the Correlation stage, and before Money Management. The order isn’t arbitrary. Filters choose the good strategies, Correlation removes the ones too much alike, and only then does it make sense to look at the shape of the resulting composition and correct its concentration. Rebalancing earlier would be premature; doing it after sizing would be too late.

Like the other stages, it acts at every rotation: from rotation to rotation the pool changes, the composition changes, and the Rebalancer reapplies its rule each time — recalculating averages and thresholds from scratch on the current composition. Its effect therefore isn’t about a single moment, but shapes the entire portfolio equity over time — exactly what makes it a risk-containment tool and not a cosmetic touch-up.

4. The configuration dialog

The Rebalancer opens from the ⚖️ Portfolio Rebalancer button in the Portfolio Rules header. It’s disabled by default: it must be explicitly enabled. The dialog is organized into four blocks.

4.1 Activation

Soft mode is a valuable analysis feature. It lets you see how much and which concentration the Rebalancer would remove, without changing anything. It’s the ideal way to understand whether your portfolio is more unbalanced than you thought, and to tune thresholds, before switching to Hard and letting it actually intervene.

4.2 Dominance Controls

Here you set the dominance thresholds, at two independent levels you can enable separately. Each threshold is a multiplier (field with a x suffix), not a maximum number of strategies.

You can use one level, the other, or both. The two controls work together: a strategy can be removed because its ticker is disproportionate, or because its asset class is. When several groups overshoot at once, the Rebalancer attacks first the one most out of scale relative to its own threshold.

How to read the threshold. 2.0x means “at most double the average of the others”. The field accepts values from just above 1.0 upward: a threshold of 1.0 or less would make almost every group dominant (being just above the average would be enough), and would no longer represent an accepted imbalance. That’s why allowed values start at 1.1x.

4.3 Activation Conditions

Conditions that keep the Rebalancer from acting when it wouldn’t make sense.

Why “Min. distinct groups” matters. The Rebalancer’s rule is based on a comparison: a group is dominant relative to the others. If at a given level there’s only one group — for example all surviving strategies belong to the same asset class — there’s no “other” to compute an average against, and the question loses meaning. In that case the level is simply skipped, without forcing arbitrary exclusions.

Why protecting a ticker from zeroing out matters. The Rebalancer’s purpose is to diversify, not to impoverish. If, to come within a threshold, it ended up wiping out an instrument’s presence entirely, it would end up reducing diversification instead of increasing it. The protection, active by default, holds both goals together: lightening disproportionate groups without emptying out the ones that give the portfolio variety. When reducing a dominant group would require zeroing out a ticker, the Rebalancer prefers to stop and leave that group slightly above threshold.

Why the “weight of other groups” matters. The Rebalancer reduces the leading groups by comparing them to the rest of the portfolio. But if that rest thins down to a single tiny group, using it as a reference would mean dismantling most of the portfolio chasing a proportion dictated by a detail. This threshold requires the “other” groups to carry sufficient weight — by default at least one third of the groups — before letting rebalancing act. It’s also what keeps the Rebalancer from pushing the composition toward forced uniformity: when the rest gets too thin, it stops.

4.4 Exclusion Criteria

When a group is dominant, a strategy must be removed: this section decides which. It’s an ordered list of criteria, reorderable by drag & drop (or with up/down buttons), and the rule is clear: the strategy with the lowest value according to the first criterion in the list is excluded. In case of a tie, the next criterion acts as a tiebreaker.

Available criteria are Net Profit, Profit Factor, Sharpe Ratio and Avg Trade; the default order favours Net Profit, then Profit Factor, then Sharpe Ratio. The order is what matters: putting the criterion that best represents “quality” for you at the top means always sacrificing the strategy that, by that yardstick, contributes least.

The PR Exclusion criteria override. The criteria set here can be overridden by those defined in the Portfolio Rebalancer — Exclusion Criterion box of the Portfolio Rules tab. If you’ve entered one or more criteria there (for example a Custom Index synthesizing several metrics into a single quality score), those take precedence for that generation, and the dialog flags it with a banner. It’s the most powerful way to steer the Rebalancer: instead of the four standard criteria, you give it a custom composite judgement. The PR Exclusion box is described in the main Portfolio Rules chapter.

5. The algorithm, step by step

Under the hood the Rebalancer works iteratively and deterministically:

1. Count strategies for each active group (ticker and/or asset class).
2. For each level with enough distinct groups (at least Min. distinct groups), identify the leading groups (those with the maximum count) and treat groups below them as “others”. If the “other” groups don’t reach the required minimum weight (Min. weight of “other” groups), the level is skipped; if there are no “other” groups — all groups tied — the composition is already balanced and no action is taken.
3. Compute the average count of the “other” groups and the limit threshold × average. If the leading groups exceed it, they’re dominant. When several levels have dominant groups, attack first the one most out of scale relative to its own threshold.
4. Remove one strategy from each leading group (one if there’s a single leader, one per co-leader when several are tied): for each, identify the worst according to the exclusion criteria (with the cascading tiebreak seen above), skipping ones that would zero out their ticker if the protection is enabled.
5. Start over: counts change, the set of leading groups and the “others” average are recalculated, and with them the effective limit. The cycle continues until no group exceeds its threshold anymore, or until the “other” groups quorum is no longer met — stopping in any case before dropping below Min. strategies in portfolio, and never zeroing out a ticker when the protection is enabled.

The process is repeatable: given the same input it always produces the same result, regardless of the order strategies arrived in. For each exclusion the Rebalancer also logs the reason in readable form — for example “over-represented ticker: 8 strategies vs cap 6 (threshold 1.5x on others’ average of 4.0)” — useful for understanding after the fact why a given strategy dropped out.

6. Common issues and notes

I enabled the Rebalancer but it excludes nothing. Several possible causes, all legitimate: the portfolio has fewer strategies than Min. strategies in portfolio; no group exceeds its own dominance threshold; the level has fewer distinct groups than Min. distinct groups and is skipped; or you’re in Soft mode, which computes exclusions but doesn’t apply them. Check the mode, the thresholds and the number of groups.

Exclusions are computed but the portfolio stays the same. That’s Soft mode’s behaviour. Switch to Hard to actually apply them.

I change the criteria in the dialog but they’re ignored. You probably have criteria set in Portfolio Rules’ Exclusion Criterion box: those override the dialog’s criteria for generation. The banner in the dialog flags this.

I lowered the threshold but the portfolio barely thins out. The Protect ticker from zeroing out protection and the Min. strategies in portfolio threshold act as brakes: the Rebalancer never zeroes out a ticker or drops below the minimum, even at the cost of leaving a group slightly above its threshold. It’s a design choice that puts diversification ahead of rigidly hitting the exact proportion.

Only one asset class among the survivors and the asset-class check doesn’t trigger. That’s correct: with a single group there’s no “average of the others”, and the level is skipped (see Min. distinct groups). The per-ticker check can still stay active if there are enough distinct tickers.

Two groups are tied at the top but the Rebalancer doesn’t touch them. The “other” groups probably don’t reach the minimum weight (Min. weight of “other” groups): if too little remains below the two leaders — for example a single tiny group — rebalancing abstains, to avoid dismantling the portfolio chasing a proportion dictated by a detail. Lower the quorum threshold if you want it to intervene in these cases too, keeping in mind that doing so pushes the composition toward greater uniformity.

7. An example workflow

An exploration path, not a recipe. The right thresholds depend on your pool and your diversification goals.

1. Open ⚖️ Portfolio Rebalancer from the Portfolio Rules header and enable the module.
2. Start in Soft mode to observe what the Rebalancer would remove, without changing the portfolio.
3. Set the thresholds in Dominance Controls: how disproportionate a ticker and/or an asset class can be relative to the average of the others (in multiples, e.g. 2.0x).
4. Decide the exclusion criteria (or let a Custom Index from Portfolio Rules’ Exclusion Criterion box dictate them).
5. GENERATE PORTFOLIO and observe, in the Equity tab and the analysis tools, the effect on concentration and equity smoothness.
6. When the thresholds convince you, switch to Hard mode to actually apply the rebalancing, and regenerate.

The Portfolio Rebalancer is a discipline tool: it translates the intent of not putting too many eggs in the same basket into an automatic rule. But you write the rule, and the effect on the portfolio’s real risk should always be verified — with Walk-Forward Analysis, stress testing and drawdown observation — before considering it settled.

8. To learn more

• Portfolio Rules — the rule chain, and the Exclusion Criterion box that can override the Rebalancer’s criteria.
• Correlations — the decorrelation stage that precedes the Rebalancer.
• Money Management — sizing, which acts downstream on surviving strategies.
• Custom Index — for building a composite exclusion criterion.
• Tools Tab — instrument classification and asset classes.