AI Portfolio Analysis: Which Holdings Moved the Return?

A portfolio can rise 1% and still trail its benchmark. In the example below, one holding contributes 200 basis points, another takes away 120, and the fund underperforms by 40. That calculation identifies where to investigate. Explaining the moves requires a second layer of work: company evidence, market context, and the portfolio's original thesis. AllMind can help connect that research to the holdings that actually affected the result.

A $100 million portfolio, fully reconciled

Take a deliberately simple portfolio: three holdings and cash, no trading during the period, no external flows, and all returns measured in the portfolio's currency. The names and returns are illustrative. They are not client positions, market observations, or output from an AllMind run.

The closing value is $101 million. Dividing the $1 million gain by the $100 million opening value gives a 1% portfolio return. The contribution calculation gives the same answer: 200 minus 120 plus 20 equals 100 basis points.

For each position, multiply its opening weight by its period return. Holding B's 30% weight multiplied by its −4% return is −1.2 percentage points, or −120 basis points. Its own return and its contribution are different quantities. Mixing them produces a persuasive explanation of the wrong number.

Suppose the benchmark returned 1.4% over the same period. The portfolio underperformed by 0.4 percentage points, or 40 basis points. Holding A was still the largest positive contributor. A profitable holding can contribute to a fund that underperforms.

This is single-period contribution arithmetic. Trades, fees, derivatives, currency movements, and external cash flows change the calculation. Use the firm's official performance record for those cases; a language model should not improvise the accounting method. The GIPS standards for firms are a reference for professional performance presentation, not a certification of this example.

The largest detractor is only the start of the explanation

The table establishes that Holding B cost the portfolio 120 basis points. It does not explain why B fell 4%, whether the position was a mistake, or whether the investment thesis has weakened.

Those questions need different evidence. A return calculation needs positions and prices. A company explanation needs disclosures and context. A judgment about the position needs the original thesis and the risk the manager intended to take.

Timing rules out some attractive stories. If B published disappointing earnings after the final market close in the measurement window, that announcement cannot explain the return already recorded during the window. It can matter to the next period and to the thesis, but it belongs in a separate part of the note.

A peer comparison can also change the question. If B fell 4% while genuinely comparable companies fell 7%, “weak company news caused underperformance” is not established by the price chart. The investigation should first determine why the group fell and whether B's relative resilience has an identifiable explanation. Relative performance provides context; it does not prove causation either.

Keep the wording proportional to what is known. A filed guidance cut before the sell-off establishes a relevant event. Proving how much of the price move it caused is a harder claim. If several events overlap, the note should identify them and explain which account has the strongest support.

A winning position can deserve the next research hour

Holding A ended the period at $42 million. Its closing portfolio weight is therefore $42 million divided by $101 million: approximately 41.6%, up from 40%, even though nobody bought more shares.

That gives the manager another useful question. Has the fund's exposure grown beyond the intended position size? The answer depends on the mandate and the thesis, which the return table does not contain. But the arithmetic has identified a decision worth making.

A generic “winners and losers” summary would probably praise A and focus the investigation on B. A better review distinguishes three priorities:

• Performance contribution: A added the most and B detracted the most.
• Changed exposure: A now represents a larger share of the portfolio.
• Changed investment evidence: any holding may deserve attention if a new disclosure alters a central assumption.

The third priority can involve C, despite its modest 20-basis-point contribution. A small price move does not establish that nothing important happened to the business. That is why a performance review needs both the calculated return and the underlying research.

Give AI the contribution table and the old thesis

Start the research with the approved table, the exact measurement window, and the manager's existing view. Otherwise, the system is being asked to infer which news matters without knowing the positions, the benchmark, or the reasons for owning them.

For B, the first assignment is specific: identify disclosures released within the period that bear on the thesis, compare them with the previous guidance, and distinguish company news from a shared industry move. A useful answer opens the relevant passage and explains what changed. Repeating the day's headlines does not complete that assignment.

For A, the assignment is different: investigate whether the evidence justifies the larger exposure created by appreciation. The price gain is already known. More paragraphs celebrating it will not help decide whether the weight remains appropriate.

We recommend AllMind when this second layer repeatedly spans public disclosures and the firm's own research. An AllMind Data Room can hold the approved thesis and selected documents, while Grids can run the same research questions across the relevant companies with citations in the answers. That makes the evidence comparable without treating every holding as the same story. The official performance calculation remains an input to this work.

The note should make the next decision easier

For the constructed portfolio, a useful conclusion is already possible: A supplied 200 basis points, B removed 120, C added 20, and the fund returned 1% against the assumed benchmark's 1.4%. A's weight rose to approximately 41.6%. B warrants investigation as the largest detractor; A warrants a position-size review. No company-level cause has been established by those calculations.

Once the source review is complete, add the strongest supported explanation and the consequence for the thesis. If the evidence is inconclusive, identify the unresolved question. The manager should be able to see which number is settled, which explanation is supported, and which decision remains open.

For the related workflows, see monitoring portfolio companies, testing an investment thesis, and refreshing macro research.

Method: the portfolio and benchmark are constructed examples; all arithmetic is shown. Product descriptions refer to AllMind's linked public documentation, checked November 10, 2026. No client portfolio, product benchmark, or investment outcome is presented.