What a Reversal Must Write Down Before It Trades
When a position returns after being deliberately removed, the record should show what reopened the case, when the first trade appeared, and how the new conclusion became this exposure.
By the team at Banyantree Investment Group
A practice note developed from the record reconstructed in Three Stopped Calls Before a US$6bn Technology Purchase.
In January 2000, Stanley Druckenmiller's decision to hold less technology had a documented case behind it: valuation, deteriorating market breadth, and a structure designed to survive further gains without repeating the losses of an outright short. By March, that answer deserved another hearing. The technology specialists he had hired were still succeeding, and his momentum signal had changed direction.
The surviving record is unusually detailed on what made the old answer difficult to carry. It preserves the widening performance gap, the repeated meetings, the three calls to buy that he began and stopped, and Druckenmiller's own account of needing to participate. It is much thinner on how the reconsideration became approximately US$6bn of technology exposure.¹ The companion feature reconstructs that record. This note takes the test it performs by accident, dating every reason against the trade it preceded, and turns it into something an institution can run on purpose.
An investment organisation usually possesses the documents a later reader of that story lacks. It has the original paper, the research updates, the first proposed transaction, the risk discussion, the approval, and the orders. Yet the final memorandum often overwrites the route by which the portfolio arrived there. Later evidence is folded backwards into an earlier proposal, and the transaction appears as the natural result of work that, in reality, arrived after the desire to act.
Nor is the gap a solo manager's problem. The pressure in the case is institutional before it is personal: it arrives through performance tables, client questions, and the rising cost of explaining an old decision, and those channels reach committees as reliably as they reached one desk. An approval chain can record everything and still lose the route, because the ordinary papers rarely ask, in one place, what changed and when.
A decision-reopening record preserves that route.
It has two parts. The front sheet states the answer currently in force. Behind it sits a dated evidence and proposal ledger, retaining the old answer, every material reason for reopening it, the first action proposed, later versions, challenge, approval, and execution.
A portfolio that cannot change its mind cannot learn, and the record is built accordingly. Its object is to make the change reconstructable before the profit or loss supplies a cleaner explanation.
Start with the answer already in force
A reconsideration should begin by retrieving the paper the portfolio is attempting to replace.
That paper may concern a security, a manager, a hedge, an asset allocation, or a decision to hold nothing. It should identify the conclusion, the portfolio role, the intended exposure, the main assumptions, the horizon, and the people authorised to reopen or change it.
It also needs a field investment papers often omit:
Accepted carrying costs: what may become painful without, by itself, invalidating this decision?
The answer might include relative underperformance, negative carry, tracking error, lost income, an insurance premium, delayed confirmation, repeated client questions, or a period during which the position appears unnecessary.
State how much pain was contemplated and for how long. Then state what would turn that pain into evidence requiring review.
A defensive position might be expected to cost 40 basis points a year in ordinary markets. A widening cost beyond 80 basis points, caused by a change in the hedge's behaviour rather than a general rally, might trigger reconsideration. An underweight may be expected to detract while valuation remains extreme. Stronger earnings, better balance-sheet evidence, or a different market structure could reopen the conclusion; the benchmark continuing to rise may already have been accepted.
Without this field, ordinary discomfort returns later wearing the appearance of new information.
The original paper should also record its reopening signposts. These are not a closed list. The world can change in ways the analyst did not predict. But signposts written before the pressure arrives establish which developments the institution already considered capable of changing its mind.
Where the original paper is missing, reconstruct it before writing the new one. Mark the reconstruction as retrospective, identify the documents and people relied upon, and separate what was recorded then from what is remembered now. An imperfect old answer is more useful than a polished fiction about one.
Use the record where a decision changes why an exposure is held, the risk it may impose, the period for which it will be carried, or the authority under which it remains in the portfolio. Re-entering a deliberately exited exposure, removing a hedge bought for a stated purpose, lifting an established boundary, extending a thesis after its expected timing has failed, or relabelling a tactical holding as strategic would ordinarily qualify. Renaming is how reversals travel in disguise: the security may be unchanged while the portfolio decision has reversed. Routine rebalancing can remain in routine records.
Give every reason its proper name and date
The question "What changed?" is too broad. Three different things may have changed, and each belongs in the record under its own name.
Investment evidence concerns the asset, strategy, manager, or market: earnings, valuation, competitive position, capital structure, liquidity, price behaviour, or information that changes the probability or magnitude of the original scenarios. Each entry should identify its source, the original assumption affected, and the revised implication.
Portfolio conditions concern the owner: available risk capacity, liabilities, redemptions, collateral, mandate terms, correlations, cash requirements, concentration, tax, or a change in what the portfolio now needs the position to do.
Institutional pressure records dated events, because the method has to be honest about what a room can collect. Nobody minutes their own envy. What a room can record are occurrences: relative performance, the third consecutive meeting at which the same polite question was asked about the same absent asset, the quarter the peer table started arriving stapled to the papers, a consultant review, the departure of the decision's original sponsor, the point at which the explanation began costing the person giving it. This is a confession made by chronology rather than by introspection, and chronology is the kind an institution can actually produce. Pressure recorded this way admits nothing about motive; it is part of the decision environment, and leaving it out removes only the institution's ability to examine its influence later.
The categories can overlap. A redemption warning begins as pressure when it is only a conversation. Once the amount and timing alter available capital, it becomes a portfolio condition. A rising price may be both an institutional burden and investment evidence where market behaviour formed part of the original analysis.
Record both dates where they differ:
Knowable date: when the information became reasonably available to the organisation.
Awareness date: when the relevant decision-maker received or recognised it.
Suppose an issuer files a refinancing agreement before the market opens and the analyst reads it at midday. The filing carries the knowable date. Midday is the awareness time. The distinction prevents a public fact rediscovered after a poor month from acquiring false novelty. It also shows where information existed inside the organisation without reaching the person authorised to act.
An old fact can acquire a new implication. A known customer concentration may matter differently after the customer weakens. The record should identify both: the old fact and the later development that changed its relevance.
"No material change" is also a valid entry. A reopened decision need not end in a reversal. The organisation may investigate and retain the old answer, alter the implementation, stage the trade, reduce the size, add a hedge, or postpone the decision pending evidence that has not yet arrived.
Keep the first proposal
The final investment paper is usually the best paper. It has benefited from more time, better questions, and evidence discovered during the review.
It is also capable of hiding the process that produced it.
Keep the first proposed action exactly as it appeared: its date and time, the security or instrument, the direction, the proposed size, the intended timing, the reasons cited, and the evidence available at that point. Retain every material version after it.
The first proposal deserves no suspicion for being first. What the record needs from it is a date, because a fact discovered on Friday cannot explain why a 4% position was proposed on Tuesday. The fact may justify the proposal afterwards, reduce it, enlarge it, or replace it with a better one. The chronology should show which occurred.
This is one defence against evidence farming. A capable team can find dated support for almost any trade it already wants; in 2023 the release of ChatGPT was available as a justification to every chaser of every chart in the sector. The record answers by requiring every later fact to identify:
1. the original assumption it changed;
2. when it became knowable;
3. when it entered the work;
4. and how the proposed action changed after it arrived.
Evidence that leaves the proposal untouched may still be relevant. But the institution can see that the transaction preceded it. The page raises the price of self-deception without abolishing it.
The revised conclusion and the resulting exposure should then be written separately.
The conclusion might be:
The issuer's refinancing risk has fallen enough for the bonds to become investable.
That conclusion does not decide which bond, what weight, whether entry should be staged, how much liquidity is available, what correlated exposure already exists, what downside remains, or which development would close the position again.
The exposure should bear a recognisable relationship to the evidence that reopened the question. Research into individual companies may justify owning those companies and nothing wider. A price signal may justify a tactical position with conditions attached. A changed portfolio need may justify a hedge whose size comes from the liability rather than from expected return.
Record who approved the revised interpretation and who authorised the exposure. Where the same person performs both roles, state the overlap. Small firms cannot always separate every chair. Visibility still matters. The page should show who proposed, who challenged, who approved, and whose performance or responsibilities were affected by the result.
Urgency does not remove the need for a record. Where delay would materially harm the portfolio, complete a minimum entry before trading: the answer in force, the new fact, the immediate action, the maximum exposure, the authority relied upon, and the expiry of the urgent approval. Complete the ledger promptly afterwards and mark every retrospective field.
Run the awkward case
Consider a constructed global high-yield mandate.²
On 8 January 2025, the portfolio sells its position in an issuer's 2029 bonds. Gross leverage is 5.7 times, free cash flow is negative, and a large refinancing falls due within 18 months. The bonds trade at 96.
The paper records a zero weight and 3 reopening signposts: completion of a proposed asset sale with proceeds applied to debt, gross leverage below 4.5 times, or a refinancing that materially extends the maturity.
It also records the accepted carrying cost. The bonds may rally while the portfolio waits. The mandate may underperform the benchmark by as much as 60 basis points over 6 months without that result alone invalidating the decision. A rally supported by completed deleveraging or materially improved refinancing capacity would require review.
By early April, the bonds trade at 101. The issuer has a 3.8% benchmark weight. The underweight has cost the portfolio 45 basis points. A client asks why the fund remains outside a security that competitors own.
Those events enter the ledger as institutional pressure. The price may also contain investment information, but the original paper already expected some rally and set signposts for distinguishing price from a changed credit case.
On 12 April, the first proposal appears: rebuild a 4% position immediately. The cited reasons are stronger market access, positive momentum, and the cost of remaining underweight.
The proposed trade stays in the record.
At that date, the asset sale has not completed. Debt has not been repaid. Leverage remains above the original threshold. The 4% proposal existed before those developments.
On 16 April, the issuer announces that the asset sale has closed. The announcement is public at 8.20 am and reviewed by the analyst at 10.05 am. Both times enter the ledger. On 19 April, the company uses the proceeds to repay its term loan. At month-end, new accounts show gross leverage of 4.4 times and positive free cash flow.
The investment evidence now supports reopening the zero-weight conclusion. Two of the original signposts have been met.
It does not automatically support the first proposal.
The portfolio already holds 10% in economically related transport and logistics issuers against a 12% internal group boundary. A 4% position would exceed that capacity. The bonds also trade less frequently than they did at 96, and the recovery available from current prices is smaller.
The revised conclusion is:
The credit has improved enough to own, subject to refinancing progress and the portfolio's remaining group capacity.
The alternatives considered are a 1% initial position, a 2% staged purchase, a credit-default-swap exposure, and continuing to wait for the refinancing.
The authorised action is a 2% position accumulated over 5 trading days, with no increase until the maturity is extended. The credit lead approves the revised interpretation. The portfolio manager and risk chair approve the size. The first 4% proposal remains visible in the ledger.
What the chronology shows is more useful than motive, which it cannot reach. The desire for 4% existed before the strongest evidence arrived. The later evidence justified changing the old answer. The portfolio's capacity and the remaining credit risks justified a smaller exposure than the first proposal. Reconsideration was legitimate; the ledger shows how the institution prevented reconsideration from choosing its own size.
The front sheet can now fit on one page:
1. Current answer. Own 2% of the 2029 bonds, entered in stages; no increase before the refinancing is extended.
2. Prior answer and accepted carrying cost. Zero weight from 8 January 2025; up to 60 basis points of relative underperformance over 6 months accepted while the credit signposts remained unmet.
3. What changed. Asset sale completed; debt repaid; leverage declined to 4.4 times; positive free cash flow. Knowable and awareness dates retained in the ledger. Benchmark loss and client question recorded as pressure.
4. Proposal history and alternatives. First proposal: 4% on 12 April, before the strongest credit evidence arrived. Alternatives: 1%, staged 2%, derivative exposure, or wait. Full versions retained.
5. Exposure and authority. 2% set by remaining group capacity, liquidity, and refinancing risk. Interpretation approved by the credit lead; exposure approved by the portfolio manager and risk chair.
6. Review terms. Review on refinancing, leverage above 4.8 times, deterioration in free cash flow, or a breach of the group boundary.
The page states the answer. The ledger preserves how the answer was reached.
Read the page before the result
A decision-reopening record should be reviewed without first seeing what the trade made.
Read the original paper. Read the reopening sheet. Read the evidence and proposal ledger. Only then open the performance report. Test whether the evidence was accurate, whether it affected the assumption claimed, whether the revised conclusion followed, whether the exposure followed from that conclusion and the alternatives, and whether the agreed decision rights were used.
Reversal and abstention fail differently, and review cultures punish only one of them. A failed reversal has a date, an order, and an owner. A failed decision to do nothing erodes performance and institutional support over years without producing one moment anyone has to approve, and left unguarded, that asymmetry teaches every manager in the building to capitulate quietly and call it flexibility. The record is the correction, and it protects the reverser as much as the fund: a manager whose evidence field was full, dated, and challenged can be wrong in peace, and a manager who held through the accepted costs can point to the paper that named them in advance.
For that protection to hold, the review grades the page against the information available at its date. Later evidence enters the ledger under the date it became knowable. Profit and loss enter under their own dates. Neither is written back into the earlier record.
This method cannot inspect motive. It cannot prevent a strong team from assembling a convenient case, ensure that pre-agreed signposts remain relevant, or guarantee that a well-recorded decision will be correct. A challenger may know less than the portfolio manager and carry a bias towards the status quo. A preserved first proposal, graded later against facts it could not have known, can make people slower to propose at all. Signposts age. Process can preserve a bad answer as easily as a good one.
Judgement remains with the people responsible for the capital. What the record changes is what the institution will be able to say afterwards: what it knew, what it expected to hurt, what reopened the decision, when the first trade appeared, and why that evidence authorised this exposure rather than another one. Much of the ledger draws on records a larger institution already keeps; a smaller one will find some of it new work, beginning with giving the first proposal a written form at all. In both, the discipline is preserving sequence and making the route between decisions explicit. What the page asks for is rarer, and it is the part no instrument can supply: the willingness, later, with the outcome known and the room listening, to read back its own handwriting.
General information only. Not personal advice. Calculations are illustrative. The worked example is constructed and does not describe a Banyantree portfolio, recommendation, or position. Past performance is not indicative of future performance. This material is intended for wholesale and professional investors.
Related Featured Essay
For the full account behind this practice note, read Three Stopped Calls Before a US$6bn Technology Purchase, which reconstructs the competing evidence, performance pressure, stopped calls, and approximately US$6bn technology purchase from which the decision-reopening record was developed.
Notes
1. The chronology of the March 2000 record and its complete source apparatus sit in the companion feature. The public record discussed there does not establish that investment, valuation, liquidity, or position-sizing work was absent inside Soros Fund Management in March 2000.
2. The high-yield example is constructed to demonstrate the decision-reopening record. The issuer, prices, leverage, dates, benchmark weight, limits, proposals, and approvals are illustrative, and the sequence is tidier and the dates more compressed than practice supplies. They are not forecasts, recommendations, Banyantree holdings, or descriptions of an actual mandate.
