Crisis Performance and Path Dependence

Define the event before evaluating it

The word crisis can refer to an abrupt price discontinuity, a sustained decline, an inflation shock, a funding disruption or a combination of events. These are not interchangeable experiments. A method that uses trailing information encounters each path differently. Before discussing crisis behaviour, specify the event window, the reference variable and whether the question concerns direction, cash needs, drawdown or an independently documented investment result.

Selecting only memorable episodes after seeing their outcomes can produce an unbalanced narrative. A stronger comparison declares event definitions and includes less convenient paths, such as an initial shock followed by a quick reversal. It also preserves observations before the event, because they determine the method's initial state. Starting every example conveniently neutral can hide exposure that would already have existed.

Read historical research without universalising it

Time Series Momentum by Moskowitz, Ooi and Pedersen studies a specified historical multi-market construction and examines its behaviour in extreme markets. That is evidence about the authors' sample and method, not a contractual promise for future crises. Different signals, sizing conventions, costs and market coverage can change the interpretation. The paper's existence does not validate a new rule merely because both are described as trend following.

Dao and coauthors' Trend Convexity at Work analyses a connection between trend behaviour and variation measured over different horizons. This gives a theoretical reason to care about the shape and persistence of a path. It should not be compressed into the claim that any trend strategy is equivalent to immediately exercisable insurance. The mapping depends on the construction and horizon being analysed.

Compare paths with identical endpoints

Take two invented reference-level sequences across observations zero through five. Path A is 100, 96, 92, 88, 84, 80. Path B is 100, 80, 80, 80, 80, 80. Both finish 20% below their starting level. Path A spreads the movement across the sample; Path B completes it at the first transition. No actual market, trade or investment result is represented by these numbers.

Define a toy signal as the sign of the current level minus its level two observations earlier. Positive means plus one, negative means minus one, and equality means zero. The first computable signal is at observation two. Apply a further two-observation delay, so the first eligible state is at observation four. Earlier states are unavailable, not retrospectively assigned a favourable direction.

Interpret what the delay reveals

At observation four, both paths inherit the negative signal calculated at observation two. At observation five, Path A inherits another negative signal, while Path B inherits a tie from observation three because 80 equals 80. The same endpoint has therefore produced different delayed states. Path B's entire decline occurred before any eligible state in this deliberately short example; the signal cannot travel backwards to precede it.

This is a timing demonstration, not an investment comparison. Multiplying the states by reference changes would introduce additional assumptions about exposure and measurement, while contract execution would require still more detail. The correct conclusion is simply that the two paths present different information to a causal rule. A claim about protection needs a separately defined and supportable performance framework, not just an appealing directional chart.

Add the reversal that a smooth story omits

Extend a falling path with a rapid return toward its earlier level. A delayed negative state may remain negative while the reference recovers because it still reflects earlier information. A faster rule might reverse sooner, but can also change direction repeatedly when the path alternates. There is no parameter-free way to recognise every persistent move while ignoring every temporary reversal using only past observations.

For an educational review, preserve the same tie convention, warmup and delay across the persistent and reversal paths. Do not shorten the delay only for the more difficult scenario. Record state changes and unavailable observations before interpreting them. This exposes the trade-off between responsiveness and stability without inventing a return series or claiming that a particular speed is best for all future events.

Remember that sizing and funding follow their own paths

Even identical directions can represent different exposures when sizing responds to estimated volatility. A jump can raise the estimate after the jump has occurred, reducing subsequent size without preventing the initial surprise. A cap can also keep an otherwise rising allocation bounded. These mechanics make a direction-only comparison incomplete as a description of the economic experience of a real account.

Funding adds another timeline. Cash obligations and collateral changes can occur before a slower rule changes state. A narrative about eventual offsetting behaviour does not establish that an account can meet interim demands. Crisis analysis is more useful when directional response, risk scaling and liquidity are presented as separate questions, each with stated information availability and assumptions, rather than collapsed into a single reassuring label.

Keep the conclusion proportionate to the evidence

A credible event note distinguishes published historical findings, theoretical mechanisms and invented teaching paths. It gives each its proper evidential weight. Academic results can motivate questions; a toy sequence can verify timing logic; neither substitutes for a documented account record. When only directions and turnover proxies are available, report those outputs directly and leave investment performance claims outside the scope of the comparison.

The most useful question is not whether a strategy has a crisis label, but what sequence of information and adjustments would be required for its intended behaviour to appear. Persistent movement, sudden discontinuity and rapid reversal place different demands on that sequence. Understanding those differences makes uncertainty explicit and prevents a favourable historical episode from being presented as a guarantee of immediate or universal protection.

Source record

1. Moskowitz, Ooi and Pedersen: Time Series Momentum, NYU-hosted paper
2. Dao and coauthors: Tail Protection for Long Investors, Trend Convexity at Work