Ulcer Index¶

The Ulcer Index is the root mean square of percentage drawdowns below a running high, measuring how deep and persistent a past investment path stayed underwater.

Version: v2 · 2026-10-03 · History
Domain-specific #: 13684
Domain group: Social Sciences
Origin domain: Economics & Finance
Subdomain: Investment Risk Metrics → Economics & Finance
Aliases: UI drawdown index

Core Idea¶

The Ulcer Index (UI) summarizes the depth and persistence of historical investment drawdowns. Starting with a specified ordered series of positive portfolio values or fund net asset values, retain the highest value observed so far. At each sample calculate the percentage below that high; a new high contributes zero. Square every percentage drawdown, average over all samples, and take the square root. A deep trough weighs heavily because of squaring; a long time below the former peak adds repeated nonzero observations. The creator, Peter G. Martin, describes developing UI in 1987 and first publishing it with Byron McCann in their 1989 mutual-fund guide.^[1]^[2]

UI is a retrospective path statistic, not a forecast of future loss and not a universal definition of financial risk. It differs from return standard deviation, which counts upside and downside variation and is insensitive to their ordering. It also differs from maximum drawdown, which records only one worst peak-to-trough excursion. The creator's first-party explanation supplies both the exact calculation and two worked historical comparisons; their figures should be read as historical demonstrations, not investment recommendations.^[1]

Structural Signature¶

Sig role-phrases:

Positive valuation path: ordered observations of the same fund or portfolio, ideally with comparable reinvested-distribution and cost conventions.
Running high-water mark: the greatest value from the beginning of the analyzed series/window through each observation.
Percentage drawdown: the current shortfall from that mark, zero at or above a new high.
Square-mean-root aggregation: depth-weighted average over every observation, so time underwater matters.
Window and comparison convention: the period and sampling frequency, plus consistent treatment of dividends, fees and risk-free benchmark if UI is used in a return-per-risk ratio.^[1]

For value V_t>0 and within-window running maximum M_t=max(V_1,..,V_t), define d_t=100×(V_t/M_t−1)≤0 and UI=sqrt((1/n)Σ d_t²). Squaring removes the sign, but the drawdown construction first excludes upside moves. This is the author's pseudocode expressed algebraically.^[1]

What It Is Not¶

UI is not price volatility in both directions. A sequence of new highs can vary rapidly upward while each d_t remains zero. It is not the single maximum drawdown, because all underwater observations count. It is not a frequency of losing periods: a portfolio may have few negative returns yet remain below an old high for a long time. It is also not the Ulcer Performance Index (Martin ratio): that separate ratio divides excess return by UI, using UI as the risk denominator.^[1]

The high-water-mark convention matters. Martin's pseudocode updates one MaxValue from the start of the analyzed sequence as it walks forward. It does not replace that high at every observation with a separate sliding last-n-bar high that can forget an older peak inside the same calculation. Tools may offer rolling calculations, but each reported UI needs its own defined window and peak-reset rule. Sample frequency matters too: a quarterly series can miss a drawdown and recovery that weekly values reveal. Historical UI is not a probability of future loss.^[1]

Scope of Application¶

Martin devised UI for mutual-fund comparison and presents it as a way to ask what a long holder experienced when an investment fell below a prior peak. His first-party article shows monthly 2000–2009 values of an S&P 500 index mutual fund with dividends reinvested. The actual return series is rearranged into a worst-months-first sequence and a flattened sequence. The three paths use the same monthly return multiset, so they retain the same annualized return (−0.52%) and monthly standard deviation (4.66%) in his report, yet their drawdown profiles differ. They are counterfactual reorderings of one path, not three separate actual funds.^[1]

The article also gives a 1940–1997 retrospective comparison of S&P 500 buy-and-hold with a simple momentum-timing backtest. Its table reports UI 8.85 versus 5.14, annualized returns 12.59% versus 14.79%, and annualized standard deviations 16.10 versus 13.18, respectively. The timing strategy's reported lower UI shows the metric's sensitivity to historical downturn avoidance in that calculation. It does not establish an executable future edge, net-cost robustness beyond the stated assumptions, or suitability for any investor.^[1]

Clarity¶

Take an author-constructed five-observation value path 100, 90, 90, 100, 110. Its running peaks are 100, 100, 100, 100, 110; percentage drawdowns are 0, −10, −10, 0, 0. UI is sqrt((0+100+100+0+0)/5)=sqrt(40)≈6.32 percentage points. If the 90 appeared only once before recovery, there would be one rather than two squared 10-point observations and a lower UI. This is a formula illustration, not a source-reported market series.^[1]

Now distinguish two common summaries: maximum drawdown of that path is 10%, indifferent to whether one or many observations stay at 90; return standard deviation counts all returns including the final rise to 110. UI specifically weights the negative distance from a prior high at each observation. It can therefore separate histories that share the same return distribution after reorder—as Martin's 2000–2009 figure demonstrates.^[1]

Manages Complexity¶

The statistic compresses a sequence of peak, decline and recovery into one number without discarding order. Squared underwater values carry both severity and duration through repeated samples. The 2000–2009 constructed reorderings make this visible: return standard deviation and total return stay the same because the monthly returns are only rearranged, while the peak-relative paths and their UI differ. That is precisely the information the risk measure was designed to retain.^[1]

Compression also loses information. UI does not tell which date the trough occurred, what caused it, whether losses were realizable at shown prices, how investors behaved, or what the next downturn will be. It ignores gains at new highs, deliberately. A single UI computed on one historical regime cannot certify future downside safety. Martin's own caution asks for long and common comparison periods, bull and bear markets, reinvested distributions and net costs; those conditions are necessary for a fair retrospective comparison but not sufficient for prediction.^[1]

Abstract Reasoning¶

Fix an observation window and frequency, then initialize the running maximum with the first positive value. For each t, update that maximum if V_t exceeds it; compute d_t as the percent shortfall; include zero for a high; finally compute the root mean square. Two paths with the same return multiset can have different UI because the running peaks depend on order. Two paths with the same worst trough can also have different UI because the number of underwater samples differs.^[1]

To compare assets or strategies, hold period, sampling, valuation convention and treatment of dividends/fees fixed. If the use case is a return-per-risk ratio, compute the excess-return numerator separately and call the result UPI or Martin ratio, not UI itself. Do not infer that a low UI implies high return: a nearly flat investment can have low drawdown and poor performance. Likewise do not equate a high UI with a high probability of future loss; the calculation contains historical observations only.^[1]

Knowledge Transfer¶

The formula can be applied to mutual-fund NAVs, stock total-return values, commodities or strategy equity curves if the series is positive and conventionally comparable. The numerical result does not transfer across unlike sample periods or data frequencies. A weekly calculation can see drawdown-and-recovery episodes that quarterly sampling misses; the creator explicitly discourages coarse quarterly-or-longer intervals for that reason. Comparing a dividend-reinvested fund with a price-only stock would also mix return conventions.^[1]

The creator's S&P 500 timing table illustrates one historical application, not a general rule that timing beats buy-and-hold. The underlying principle that transfers is peak-relative depth integrated over observations. Investors with short positions or different utility functions may not regard upside variation as harmless; in those contexts standard deviation or other measures may answer a different question. UI remains one lens, not a comprehensive risk model.^[1]

Cross-Domain Echoes¶

See how this entry connects to another domain.

The rule decides what dominates the summary

Examples¶

Same returns, reordered paths in the creator's 2000–2009 illustration¶

Martin takes the S&P 500 index mutual fund's monthly 2000–2009 return set, with reinvested dividends, and plots its actual order beside two constructed reorderings: the worst months first and a flattened arrangement chosen to reduce drawdowns. All three retain annualized return −0.52% and monthly standard deviation 4.66%, since they use identical monthly returns. Their underwater shapes differ. The figure is an executed order-sensitivity demonstration, not a claim that three real funds generated separate histories or that an investor could know the best order ahead of time.^[1]

Mapped back: monthly NAV paths are the positive valuation paths; rearranging returns changes each path's running high-water mark; the below-high distances are distinct percentage drawdowns; squaring and averaging those sequences yield different UI aggregation despite equal SD; the common 2000–2009 monthly data and identical return multiset are the window and comparison convention.

Reported buy-and-hold versus timing backtest, 1940–1997¶

The creator reports a long historical comparison of reinvested S&P 500 buy-and-hold with a simple momentum-timing strategy. His table lists UI 8.85 for buy-and-hold and 5.14 for timing, while standard deviations are 16.10 and 13.18, and annualized returns 12.59% and 14.79%. The reported UI decline is larger proportionally than the SD decline in this retrospective calculation, illustrating why the choice of risk metric can change a risk-adjusted comparison. The strategy rules, execution costs and out-of-sample performance are not independently validated here; the figures should not be used as a trading recommendation.^[1]

Mapped back: the two strategy equity/NAV histories are positive valuation paths; each has its own running high-water marks; its below-peak values supply percentage drawdowns; the root-mean-square summaries are the reported 8.85 versus 5.14 UI values; the shared 1940–1997 period, reinvestment assumption and creator's caution on net costs are the comparison convention. An output number is not a forecast.

Structural Tensions¶

Downside path sensitivity versus symmetric return dispersion. UI gains sensitivity to how far and how long a long investment stayed below its earlier peak; it deliberately assigns zero to upside moves at new highs. The price of that focus is losing a general measure of return variation, including positive surprises and fluctuations not yet below peak. Standard deviation preserves upside-and-downside dispersion and is easily computed from return observations, but it discards return order and can equate the creator's rearranged paths despite different underwater experiences. Neither metric universally subsumes the other. The tension is between information retained by competing summaries, not a promise that one label is safer. Diagnostic: is the decision about historical time underwater, overall return dispersion, or both—and are periods and conventions identical?^[1]

Structural–Framed Character¶

UI is mathematically structural in its high-water-mark recursion and RMS formula, but its interpretation is framed by investor perspective: its originator chose long-holder drawdown discomfort as the relevant downside experience. Its evaluative weight is explicit in the name “ulcer” and in the choice to ignore upside volatility; not all financial decisions share that preference. The measure arose in mutual-fund comparison and risk-adjusted performance practice, not in medical measurement. It travels to other positive investment paths when the same valuation, sampling and peak conventions are recognized. Importing a 1940–1997 backtest's UI into a future guarantee, or using mismatched windows while claiming one fund is safer, is false transfer. Its character: a precise retrospective drawdown-path statistic whose practical meaning depends on a stated risk perspective and data convention.^[1]

Structural Core vs. Domain Accent¶

The skeletal relation is positive time-ordered value → running peak → percent underwater at each sample → squared mean → root, followed by comparison under a fixed window. The S&P 500 and a hypothetical stock are domain accents. The mechanism remains domain-bound to valuation paths and the investor's downside-risk interpretation; generic “penalize deviations” would lose the previous-high reference and exclusion of gains. The named entry fails the prime bar because its exact computational rule and financial meaning are constitutive. It presupposes the live Aggregation prime for the many-to-one summary, but aggregation does not supply the running-high rule or predict future loss. A portable peak-relative persistence prime would need genuinely unlike domains and a justified mapping beyond this metric.

This entry presupposes Aggregation.

Strict compositional prerequisite: Aggregation (presupposes, strict). UI reduces sampled drawdowns to one root-mean-square historical statistic; many aggregations lack its running-high and downside rules. Generic volatility and risk remain conceptual neighbors. The edge is not a forecast or investment recommendation, and sample interval and window still govern interpretation.

Relationships to Other Abstractions¶

Current abstraction Ulcer Index Domain-specific

Parents (1) — more general patterns this builds on

Ulcer Index presupposes Aggregation Prime

UI presupposes aggregation of sampled drawdowns.

Hierarchy path (1) — routes to 1 parentless root

Ulcer Index → Aggregation → Micro Macro Linkage

Neighborhood in Abstraction Space¶

Ulcer Index sits in a sparse region of the domain-specific corpus (66^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Financial Markets & Pricing Anomalies (13 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

UI is not standard deviation, maximum drawdown, Value at Risk, expected future loss or a medical measure. A rolling n-bar maximum that forgets an older high within the analyzed series is not automatically Martin's within-window running-high rule. UPI/Martin ratio uses UI as denominator with excess return in the numerator, so the two should not share a label. Values computed at unlike frequencies, periods or distribution/cost conventions cannot be ranked as though they were directly comparable. The creator's backtest table is evidence about those historical assumptions only.^[1]

References¶

[1] Martin, Peter G. “An Alternative Approach to the Measurement of Investment Risk & Risk-Adjusted Performance.” Creator's first-party explanation of the Ulcer Index, copyright 1987–2011, especially pp.2–5 formula, original illustrations, timing comparison and caveats: https://www.tangotools.com/ui/ui.pdf . The creator notes the index was first described in the 1989 book with Byron McCann. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u

[2] Martin, Peter G., and Byron B. McCann. The Investor's Guide to Fidelity Funds: Winning Strategies for Mutual Fund Investors. John Wiley & Sons, 1989. Bibliographic original-book record only; full book not inspected: https://books.google.com/books/about/The_Investor_s_Guide_to_Fidelity_Funds.html?id=nOoJAQAAMAAJ . registry ↩