Risk, drawdown and time to recovery

Two investments both lose 20%. The first regains its previous peak in three months. The second takes four years. The maximum loss is the same. The economic and behavioural experience is not.

Risk has at least three dimensions: how much you lose, how quickly it happens and how long the capital stays below its previous level. To these, add risks a historical series does not see well: credit, liquidity, counterparty, concentration, leverage, currency and legal structure.

A useful guide does not look for the definitive number. It teaches you to ask what each indicator measures, what it leaves out, and which data history produced it.

Six questions before reading the risk

The PRIIPs SRI: a summary, not a verdict

In PRIIPs KIDs the Summary Risk Indicator, or SRI, places the product on a scale from 1 to 7. The methodology of Commission Delegated Regulation (EU) 2017/653 combines a market-risk measure with a credit-risk class when the latter is relevant for the product.

For many instruments the market component derives from the historical or simulated distribution of returns and from the VaR-equivalent volatility provided for by the methodology. Other products follow specific categories and procedures. The credit component can change the final class depending on the structure and the exposure to an entity’s solvency.

This precision matters: saying only “the SRI measures volatility” is incomplete; saying “the SRI measures all risks” is false.

What the 1–7 scale means. A higher number signals, under the regulatory assumptions, a higher synthetic risk class. It does not mean that:

• the maximum loss is limited to a preset percentage;
• a class-3 product cannot lose capital;
• two products with the same number have the same risks;
• the number is a forecast for the next twelve months;
• the lowest class suits every investor;
• a change of class automatically implies a change in the investment policy.

The KID surrounds the indicator with warnings, scenarios, a recommended holding period and the risks not adequately captured by the number. That frame is part of the information: extracting only the digit impoverishes the document.

SRI and the old SRRI are not the same thing. UCITS documents prior to the general transition to PRIIPs KIDs commonly carried the Synthetic Risk and Reward Indicator, or SRRI — also on a 1–7 scale, but based on a different methodology, mainly tied to historical volatility. The graphic similarity does not automatically make the two indicators comparable. When a database keeps the technical name srri for historical reasons, you must check which number is actually read from the current document: the column’s label cannot replace the source.

The class can change without the product “becoming another”. The statistical window rolls, markets change regime, the document gets updated, and the methodology can produce a different class. That alone does not prove the manager changed the mandate. Likewise, a stable class does not prove all risks stayed unchanged.

Volatility: how dispersed the returns are

Volatility is normally calculated as the standard deviation of periodic returns. With daily data, annualisation often uses:

annualised σ = daily σ × √N

where N is the conventional number of periods in a year. The formula presupposes conditions that may not hold perfectly in reality: independence, a stable distribution and consistent frequency.

Volatility answers the question “how much did returns move around their mean?”. It does not directly answer “how much would I have lost?”.

It treats rises and falls alike. A strongly positive return increases dispersion just like a strongly negative one. That is why volatility is useful for comparing instability and for many risk models, but it does not coincide with the feeling of loss. Measures such as downside deviation consider only the shortfalls below a threshold, while the Sortino ratio relates an excess return to that negative deviation. They change the question; they do not remove the dependence on period, frequency and assumptions.

Volatility can arrive late. After a sudden crash, realised volatility rises because the move has already entered the window. If the market stays calm for long, the measure can fall while concentrations, duration, credit or liquidity build up in ways hard to see in recent returns. Low volatility is therefore not a synonym of absence of risk: it can describe observed stability, not guaranteed resilience.

Drawdown: the distance from the previous peak

The drawdown at time t compares the current value with the maximum reached up to that moment:

Drawdown(t) = V(t) ÷ previous maximum of V − 1

If a value rises from 100 to 120 and then falls to 90, the drawdown is not −10% from the starting point. It is:

90 ÷ 120 − 1 = −25%

The maximum drawdown is the deepest drawdown observed in the window: the largest percentage loss from a peak to a subsequent trough.

The three dates of an episode

Every complete drawdown contains at least three moments:

• Peak: the last maximum before the decline.
• Trough: the lowest point before the climb back.
• Recovery: the first date on which the value returns at least to the previous maximum.

From these dates come three different durations: the decline, from peak to trough; the recovery, from trough back to the peak; the time under water, from peak to recovery.

Writing “it recovered in 18 months” without declaring which interval is measured creates ambiguity. From the maximum loss to break-even it may be 18 months, while from the initial peak to break-even it may be two years.

An open drawdown does not yet have an end. If the value has not returned to its previous maximum, the episode is still in progress. The observed duration can be stated “up to the date”, but it must not be turned into a definitive recovery time. This detail avoids a common survivorship error: analysing only concluded episodes makes still-open losses invisible.

Why a loss requires a larger gain

To recover a loss L, expressed as a positive fraction of the capital, the required return is:

Recovery return = 1 ÷ (1 − L) − 1

The relationship is not symmetric, because the gain applies to a reduced base. After a 50% loss, the capital is halved; what remains must double to return to the initial value.

This arithmetic does not predict how long the recovery will take. It only says how much cumulative return is needed.

Time to recovery: the risk the calendar makes visible

Two ETFs with a similar maximum drawdown can have very different paths. One can fall quickly and climb back just as fast; the other can stay for years close to, but below, its previous maximum.

Time to recovery matters because capital can have an economic deadline. A temporary loss is different for someone who does not need the money and for someone who must fund an expense during the downturn. The chart does not know the reader’s horizon.

Nominal time and real time. Returning to the same nominal NAV does not necessarily mean recovering purchasing power. If prices have risen in the meantime, the real recovery requires a higher level. Distributions and taxes can also change the investor’s experience. That is why you must declare whether the series is:

• price or total return;
• nominal or inflation-adjusted;
• gross or net of certain costs and taxes;
• expressed in the currency of the class, of the listing or of the investor.

The data frequency changes the photograph

A maximum drawdown calculated on monthly data can miss an intra-month trough visible in daily data. Intraday data can show an even deeper episode.

No frequency is universally correct. The choice must be consistent with the question:

• daily data to describe the path in more detail;
• monthly data for long comparisons and less noisy series;
• official NAVs to measure the fund;
• exchange prices to study the tradable experience — including, however, premium, discount and microstructure.

Mixing frequencies and sources can make the product with more granular data look riskier — not the one with the greater economic risk.

Distributions can create false drawdowns. When an ETF pays out income, the NAV or the price mechanically drops by the amount detached. A price-only series reads that decrease as a loss, even though the investor received cash. Comparing accumulation and distribution requires a total return series that reinvests the income consistently. If the distributions are not available, declaring the limit is more correct than computing an apparently precise drawdown. The same rule applies to splits, class mergers, currency changes and series replacements: before the calculation comes the semantic continuity of the data.

Correct indicators, wrong conclusions

Sharpe ratio. It relates the excess return over a risk-free rate to volatility. It is useful for disciplined comparisons, but it changes with period, currency, frequency, the chosen rate and the distribution of returns. A higher past value does not prove future superiority.

Sortino ratio. It uses a measure of negative deviation. It can be closer to the question “how unfavourable were the oscillations?”, but it requires a threshold and does not automatically capture drawdown duration, liquidity or non-linear losses.

VaR and Expected Shortfall. Value at Risk estimates a loss threshold for a horizon and a confidence level; it does not say how severe the loss beyond that threshold can be. Expected Shortfall looks at the average of the tail losses beyond the quantile. Both depend on the model and the history used.

Calmar ratio. It relates a return to the maximum drawdown. It explicitly introduces the loss from the peak, but the denominator depends on a single extreme episode in the window and can change abruptly when that episode enters or leaves the observed period.

No ratio turns risk and return into a truth without context.

Risks the historical drawdown may not show

Concentration. An ETF with many positions can depend on few issuers, countries, sectors or factors. The number of holdings does not measure economic diversification on its own.

Credit. A bond can show contained volatility until solvency is repriced. Ratings, spreads, subordination, guarantees and recovery belong to the reading of risk.

Rates and duration. For bond ETFs, duration and the shape of the curve drive sensitivity to rates. A quiet past in a stable regime does not remove the risk of a rapid change in yields.

Liquidity. The NAV can be stable while the underlyings become hard to trade. Spread, depth and price discovery were covered in chapters 3–4 and do not coincide with the NAV’s historical risk.

Counterparty and structure. Swaps, securities lending, collateral and collateralised debt securities introduce risks that may not emerge from ordinary oscillations. Chapter 1 showed why, before the return, comes the legal right.

Currency. The trading currency does not define the currency exposure. A hedged class can reduce some exchange-rate moves but introduces costs, basis and hedging imperfections. Here too, “hedged” does not mean zero risk.

From the single ETF to the portfolio

The risk of an isolated product does not say what contribution it will make to a portfolio. A volatile asset can reduce some overall risks if it moves differently from the others; an apparently stable asset can amplify losses if correlations rise during crises.

Correlation is not constant. A matrix computed in normal times can change precisely when diversification is needed. That is why the risk contribution must be observed together with weight, volatility, correlations and scenarios.

The chapter does not propose an allocation, nor does it establish which drawdown is acceptable. Tolerance depends on goals, horizon, cash flows, financial capacity and behaviour: elements a product page cannot infer.

How to read Rebalix correctly

Rebalix public fund pages already contain a section dedicated to risk, built on the NAV series declared by the issuers. The reading happens across two complementary views:

• Drawdown — distance from the previous peak: the chart shows how far the fund sits below its own maximum; the page states the maximum drawdown, the peak date, the trough date, the recovery time from trough to break-even and the overall time under water from peak to recovery. A table also lists the five worst falls of the period, distinguishing depth and duration.
• Risk indicators: annualised volatility, maximum drawdown, Sharpe and Sortino are shown for the last year, the last 3, 5 and 10 years and the full history, when the available series allows it. The currency follows the one selected in the chart; the methodology and the risk-free rate used are declared on the page.

The iShares Core S&P 500 page, for example, shows a maximum drawdown of 33.81% in the historical USD series: peak on 19 February 2020, trough on 23 March, 142 days of climb from trough to break-even and 175 days under water overall, with recovery on 12 August 2020. It is neither a backtest nor an estimate: it is an episode computed on the daily NAV series declared by the issuer. The values change when you pick another currency or time window.

Less deep can mean much longer

The same fund allows a comparison in which product, issuer, class, currency and methodology do not change:

The episode that began in 2022 was about a quarter less deep than the 2020 one, but it kept the capital under water four times longer. Depth alone would have told only half the story. Both episodes are visible in the five-worst-falls table of the same Rebalix page.

An open episode does not yet own a recovery time

The page of the Xtrackers Eurozone Government Bond UCITS ETF 1C offers the opposite case. It is an accumulation class, with the issuer’s daily NAV series in EUR since 2007. The episode began from the peak of 11 December 2020; the trough was reached on 28 September 2023, at −22.65%.

At the photograph of 15 October 2026 the fund had not yet recovered: after 2,134 days under water, about 5.8 years, the NAV was still 16.35% below the peak. There is therefore no definitive recovery time to publish: there is only a duration observed “up to the date”.

The case gathers three lessons. First: an open episode disappears if you analyse only the concluded ones. Second: contained historical volatility does not exhaust the risk of a bond ETF; duration and rate moves can produce persistent losses. Third: since the class is an accumulation one, the NAV’s drawdown is not mechanically inflated by income payouts.

This photograph is dated and will have to be updated if the fund recovers. The historical data stays true; what will change is the episode’s status and its final duration.

The page therefore distinguishes three levels:

• Official document: Rebalix does not republish the KID’s synthetic indicator. The risk section is built on the measured series; for the indicator, the page points to the KID linked on the page, where the number lives within its frame of warnings, scenarios and recommended holding period.
• Historical measure: volatility, drawdown and recovery computed only from consistent series, with an explicit period.
• Limit: insufficient history, missing distributions, series changes, a non-comparable class, or a still-open episode.

Rebalix does not turn the 1–7 number into a “safe/risky” label, nor does it use private-app features as if they existed on the public pages. Personal rules on horizon, recurring investments or drawdown thresholds belong to the user, not to the product’s classification.

Even a correctly computed measure remains a photograph of the past. The series, the currency, the frequency, the treatment of income and the window must accompany the number: without that information, two apparently comparable maximum drawdowns can describe different objects.

A checklist for reading risk

• Document: which synthetic indicator is shown, with which date and methodology?
• Perimeter: which risks does it include, and which does it leave to the descriptive part?
• Series: NAV, price or total return?
• Income: are distributions reinvested correctly?
• Currency: which currency measures the return, and which currency risk remains?
• Period: does the history include at least one relevant difficult regime?
• Frequency: daily, monthly or intraday data?
• Drawdown: what are the peak, the trough and the depth?
• Time: are decline, recovery and time under water kept distinct?
• Status: is the drawdown concluded, or still open?
• Structure: are credit, liquidity, counterparty, concentration or leverage visible in the documents?
• Portfolio: does the product add to, or concentrate, the risks already present?

Twelve mistakes to avoid

• “SRI 2 means a maximum loss of 2%.” The scale does not express a loss percentage.
• “SRI and SRRI are the same measure.” Methodologies and documents differ.
• “Low volatility means low risk in every sense.” Credit, liquidity and structure can stay hidden.
• “Volatility and drawdown are synonyms.” One measures dispersion, the other the loss from the peak.
• “A 20% drawdown requires a 20% rise.” It takes 25% on the reduced base.
• “Recovery time always means trough-to-break-even.” It can denote different intervals.
• “If it has not recovered, the time is infinite.” It is an open episode with a duration observed up to the date.
• “Price and total return produce the same drawdown.” Distributions can change the result radically.
• “The maximum drawdown is a guaranteed future loss.” It is the worst episode of the observed window.
• “More history solves every problem.” Regimes, methodology and the product can change.
• “The listing’s risk is the portfolio’s risk.” Premium, discount and microstructure add a distinct layer.
• “An ETF that is risky in isolation always worsens the portfolio.” What counts is the interaction with the other exposures.