Risk terms in plain language

Value at Risk, and how it is checked

Value at Risk (VaR)

Value at Risk, or VaR, puts a size on an ordinary bad day. A one-day 95% VaR of 1.5% means your loss should stay under 1.5% on 19 days in 20. It is a percentile of daily losses, near the fifth-worst day in a hundred. Sentinel's standard figure is a 95% bell-curve estimate: the standard deviation of your current holdings' daily returns over up to 252 trading days, times 1.645, assuming no expected gain. Pro adds historical and simulated estimates, and its Portfolio page leads with the worse of the bell-curve and historical figures. VaR is not the worst case: it says nothing about how far the twentieth day falls.

Confidence level and time horizon

Value at Risk rests on two settings. The confidence level says how often the figure should hold: at 95%, losses should cross it about 1 day in 20, a frequency, not a promise of safety. The horizon is the period the loss covers. Sentinel's portfolio and asset pages use 1 day at 95%, and the Pro Value at Risk page adds 99% and 10 days. A 99% figure is larger because it reaches deeper into the bad days, not because the two disagree. The 10-day figure is the 1-day one times the square root of 10, about 3.2 rather than 10, which assumes the days are independent and equally volatile.

Expected shortfall, also called CVaR

Expected shortfall is the average loss on the days past the Value at Risk line. Where VaR marks the edge of the bad days, expected shortfall measures their depth, so, measured the same way, it is the larger figure. CVaR is another name for it, and Sentinel's screens use both names. It appears on Pro and above, worked out from volatility with the bell curve and assuming no expected return, at 95% and on some pages 99%, over roughly the last 252 trading days, or a rolling 30 days on the risk page's market chart. Real tails are fatter than a bell curve, so it tends to read low.

The three estimates: bell curve, history and simulation

Value at Risk has three standard estimates, which can disagree. The bell curve (parametric) assumes a normal distribution, where extreme days are rare, so it tends to understate the worst days. History replays past returns and reads the loss at the cut-off. Simulation (Monte Carlo) generates thousands of possible days and does the same. On Pro and above, Sentinel gives all three as a 1-day loss at 95% confidence, and the portfolio page leads with the worse of the first two. Its simulation redraws 10,000 days from the window's own history, so, like history, it only knows days inside the window you pick, and both need about 100 of them.

Monte Carlo simulation, and the range of outcomes

Monte Carlo simulation generates thousands of random futures and reads off the range of outcomes. Sentinel draws 10,000 paths by resampling real daily returns from the last year, with their average removed, so it assumes no expected return and shows spread, not a forecast. The bands hold the middle 50% and 90% of paths, and the 1-in-20 downside is the 5th percentile, the level only 5% of paths finish below. On the Pro plan the risk page simulates the S&P 500; on the Analyst plan the scenario page simulates your own holdings. It can only reuse days from that year, so a calm year gives a narrow range.

Lookback window and trading days

A lookback window is the stretch of past days a risk figure is calculated from. A rolling measure recalculates it daily as the oldest day drops out. A market year is about 252 trading days, not 365. Sentinel annualises volatility using the square root of 252. Its Value at Risk windows count trading days: at most the last 21 for 30D, 252 for 1Y and 1,260 for 5Y. The default is 1Y, and Pro adds a switch between the three, which changes the figure. A short window reacts faster but swings more, a long one is steadier, and since windows look backwards, each estimate lags a change in market conditions.

Backtesting a risk estimate, and breaches

Backtesting a risk estimate means rebuilding it for past days from earlier data and counting breaches, days whose loss went past it. The expected count is days tested times one minus the confidence: 5 in 100 at 95%. On Pro and above, Sentinel rebuilds each day's 95% bell-curve estimate for today's holdings and weights from about the prior 126 trading days, or up to 252 on the five-year view, and the standard Kupiec test grades the count: green if it fits, red if too high, amber if too low. The one-year view tests about 126 days, so its verdict is rough. The hypothetical performance chart against SPY tests no estimate.

The Kupiec test and the traffic light

A backtest checks a loss estimate against what happened next. At 95% confidence the loss should breach the estimate on about one day in twenty, and the Kupiec test asks whether the breach count is too far off to be chance. On Pro and above, Sentinel replays its 95% bell-curve Value at Risk over past prices of your current holdings and shows the verdict as a light: green reads Well calibrated, red reads Underestimates risk, meaning too many breaches, and amber reads Over-cautious, meaning too few. The one-year view scores about 126 days, so three to eleven breaches all stay green, and the test ignores how large each breach was.

Risk contribution (component VaR)

Risk contribution, or component VaR, is each holding's share of a portfolio's Value at Risk, which is not the same as its share of the money. A small, jumpy position that falls on the same days as the rest can carry more than its weight, while one that tends to rise when the book falls contributes a negative amount. On the Pro Value at Risk page, Sentinel multiplies each weight by how that holding moved with the book over the last 252 trading days or fewer, so the rows sum exactly to the 95% one-day headline: that is the Euler split. Those relationships come from the past and can shift.

The tail, and tail risk

The tail is the far end of the range of daily results, the rare days when the loss is much bigger than usual. Tail risk is not the ordinary bad Tuesday but the few sessions a year that do real damage. Sentinel's Value at Risk assumes a bell curve: at 95%, it is 1.645 times the daily volatility of up to 252 recent trading days. On Pro, the Portfolio page leads with the worse of that and a historical figure; the Value at Risk page adds a simulated one. Real markets have fatter tails than a bell curve, so its 99% figure is best read as a floor.

What your portfolio is made of

Portfolio weight

A holding's weight is its share of your portfolio: the position's current value divided by the value of everything you hold. Sentinel computes it as units held times the latest recorded price, over the total, and uses it for the Weight column on the portfolio page, the HHI concentration score and, on Pro, the concentration alert, which by default fires when any one holding passes 25%. Weights shift whenever prices move, so a holding that outpaces the rest takes a bigger share with no trade. Value at Risk and drawdown apply today's weights to past price moves, so they describe your current mix, not the one you held then.

Concentration and the HHI score

Concentration is how much of your money rides on a small number of positions. Sentinel scores it with the Herfindahl-Hirschman Index, or HHI: each holding's share of your portfolio by market value, squared and added up. It runs from near 0 when money is spread thinly across many holdings to 1 when everything sits in one. Below 0.15 reads as well diversified, 0.15 up to 0.25 as moderately concentrated, and 0.25 or above as concentrated. The score looks only at position sizes, not at how holdings move, so ten banks held in equal amounts score as ten separate positions even though they tend to rise and fall together.

Correlation

Correlation measures whether two holdings tend to move on the same days. It runs from +1, where they rise and fall together almost in step, through 0, where one tells you nothing about the other, down to -1, where one tends to rise when the other falls. It is measured from past daily price moves, so it describes how things have behaved. Correlations also have a habit of jumping towards +1 in a crisis, which is exactly when you wanted them apart.

Diversification benefit and effective bets

Diversification benefit is the gap between the loss you would face if every holding had its bad day at once and the loss you face because they usually do not. The Pro Value at Risk page compares the weighted sum of each holding's own 95% one-day VaR with the portfolio's over the past year, shows the saving as a percentage, and gives the diversification ratio, weighted average volatility over portfolio volatility, where 1.00 means no benefit. Analyst adds effective bets, that ratio squared, on the correlation page: twelve holdings that move together can count as three. It all rests on past prices, and correlations tend to rise in sell-offs.

Volatility

Volatility is how much a price bounces around day to day, regardless of which direction it ends up going. It is normally quoted as an annualised percentage, so a volatility of 18% describes a holding whose swings have typically been about that size over a year. A higher number means a bumpier ride rather than a worse investment, and Sentinel uses recent volatility as the raw material for one of its three Value at Risk estimates.

Factor exposure and factor decomposition

A factor is a broad force that moves many holdings at once, such as the stock market or interest rates. Exposure is how strongly your portfolio moves with a factor; decomposition works out how much of your daily movement each factor explains, and how much is down to your particular holdings. From Analyst up, Sentinel fits a year of daily returns of your current holdings, at today's weights, against four US index funds: market (the S&P 500), size (small companies relative to it), tech (the Nasdaq 100 relative to it) and rates (long-dated Treasuries). The unexplained rest shows as stock-specific, and market here means the US market.

Your book, and the universe

Your book is everything you hold, taken as one portfolio. The universe is every asset tracked and compared. In Sentinel it is the shares, funds, indices, currencies, futures and crypto priced in the last 45 days with at least 60 daily returns, each with a volatility-based 1-day 95% Value at Risk over up to a year. On Pro and above, your universe percentile is the share of those assets whose VaR is at or below your book's. That book figure averages your holdings' VaRs by weight, ignoring correlation, so it tends to read above your portfolio VaR. Holdings without a VaR drop out and the rest are reweighted.

How it has performed

Hypothetical performance, indexed to 100

Hypothetical performance shows how holdings would have done over a past period. Indexed to 100 means every line starts at 100, so 112 reads as up 12% and 95 as down 5%. On Pro and higher plans, Sentinel assumes today's holdings, in today's proportions, were bought at the window's start, one year by default, and left to drift without rebalancing, with SPY, an S&P 500 fund, rebased alongside. The chart starts once every holding with price history has a close, so a short history shortens the window and a holding with none is left out. It is not your track record, and past performance does not predict future results.

Drawdown and maximum drawdown

A drawdown is how far your portfolio has fallen from its highest point so far, and maximum drawdown is the deepest such fall over a period, from a peak to the low that followed it. The portfolio page shows the deepest fall within the last 30 trading days and, as worst on record, the deepest over roughly the past year. The Pro drawdown page charts that year day by day and lists up to six falls of 3% or more, deepest first. Each figure runs today's holdings, at today's weights, back through past prices, so it shows how the current mix would have fared rather than your account's actual history.

Performance attribution

Performance attribution answers which of your holdings moved your return, and by how much. Each holding's contribution is its return over the period times its share of the portfolio at the start, so a big move in a small position can count for less than a modest move in a large one, and the contributions add up exactly to the total. On Analyst, the analysis page covers the last 21 trading days, with week, month and quarter views behind it. Sentinel works back from what you hold today, so it assumes those holdings were held unchanged throughout.

Beta

Beta says how strongly a holding or a whole portfolio has moved with the wider market. A beta of 1 has moved roughly in step with it, 1.5 about half again as hard in both directions, and 0.6 more gently. On Analyst, the performance page measures your beta against SPY, an S&P 500 fund, over the window you pick, one year by default, from the daily returns of today's mix bought at its start and held, so it describes that mix rather than your past trades. The stress sandbox, on Analyst, instead lets you type a beta for any holding, which scales how hard a scenario hits it.

Sharpe and Sortino ratios, and the risk-free rate

The Sharpe ratio is the return earned above a risk-free rate, roughly what cash pays, divided by how much it bounced around. Sortino is the same calculation but counts only the shortfalls as risk, so upward swings are not penalised. On Pro, Sentinel annualises both from the daily returns your current mix would have had over the chosen window, net of a cash rate: SONIA for a sterling base, the 3-month US Treasury rate for a US dollar base, otherwise the euro short-term rate. Treynor, on Analyst, divides excess return by beta. The latest rate is used for every day, so a five-year figure assumes today's rate throughout.

Alpha and capture, measured against a benchmark

A benchmark is the yardstick a portfolio is measured against. Alpha is the annualised return left over after beta and a cash rate are allowed for. Up and down capture compare your compounded return with the benchmark's on its rising and falling days, so down capture below 1 means you fell less. Tracking error is how much your daily gap to the benchmark varies; the information ratio divides your extra return by it. From Analyst up, Sentinel measures every portfolio against SPY, a US fund tracking the S&P 500, using today's weights over your chosen window, without converting currencies. Alpha describes that past stretch, not skill or the future.

Skewness and excess kurtosis

Skewness and excess kurtosis describe the shape of returns rather than their size. Negative skew means the loss side stretches further than the gain side: frequent small gains, occasional sharp falls. Excess kurtosis measures tail weight, how often extreme days in either direction turn up, scored so a bell curve reads 0. From the Analyst plan up, the performance page shows both, computed on daily returns of today's mix held through the chosen window, using the sample-adjusted formulas behind a spreadsheet's SKEW and KURT. One violent day can swing either figure, and the volatility-based Value at Risk ignores both, so positive excess kurtosis hints it understates extreme losses.

Cost basis

Cost basis is what you paid for a holding, usually expressed as an average price per share across your purchases. In Sentinel it is optional and typed in by hand as an average cost per share, in US dollars like the last close the entry form shows. Leave it blank and the app still shows what a position is worth, but not your gain or loss since buying it. Adding shares to a holding you already own blends the costs into a weighted average only when both purchases carry one; otherwise the holding is left with no average cost. Positions synced from a brokerage carry no cost at all.

Unrealized profit and loss

Unrealized profit or loss is the gain or loss on holdings you still own, the gap between their current price and what you paid. It is called unrealized because you have not sold, so it moves with the market and only becomes real money when you do. Sentinel works it out from the latest closing price, per holding and as a portfolio total, counting only holdings with a cost recorded and noting what share of your value that covers. Costs and prices are held in US dollars and converted at the current exchange rate, so currency moves since you bought are not counted.

Stress tests and the market

Stress tests, scenarios and shocks

A stress test asks what a severe event, a past crisis or an imagined shock, would do to what you hold today, rather than the ordinary bad day Value at Risk describes. On Analyst, each Sentinel scenario is a hand-set move, or shock, for every asset class, with specific tickers, mostly funds, set separately: the 2022 rate shock cuts shares by 19%, bonds by 13% and the Nasdaq-100 fund QQQ by 33%. You can change a holding's class or scale its shock by your own beta, and concurrent shocks compound several scenarios. These are stylised estimates, not predictions: each shock lands in one step, with no knock-on effects.

Market regime

A market regime is the broad state the market is in for weeks or months at a time, which shapes how far recent history can be trusted as a guide. Sentinel sorts it into four labels from two US readings, checked in order: Recession watch when the 10-year government yield is below the 2-year, otherwise Stress when the VIX is above 28, Bull when it is below 18, and Choppy in between. The label shows on every plan's dashboard, and the full Regime tab needs Pro or above. It reflects US markets, not UK or European ones, and describes the latest daily readings rather than forecasting what comes next.

The VIX

The VIX is a single number, worked out from S&P 500 option prices, that reflects how big a move in the US stock market traders are paying to protect themselves against over the coming month. It is often called the fear gauge, because it tends to jump when shares fall sharply. Sentinel takes its daily close as one of two inputs to the market regime label: above 28 reads as Stress, below 18 as Bull, and anything between as Choppy. An inverted yield curve outranks it, so the label stays at Recession watch whatever the VIX reads.

Credit spreads

A credit spread is the extra yield a company pays over what a government pays to borrow for the same length of time. Riskier borrowers pay more, because lenders want paying for the chance of not being repaid. When worry about defaults grows, spreads widen, and credit markets sometimes show stress before share prices do. On the regime page, from Pro up, Sentinel shows the US high-yield spread, ICE BofA's measure for lower-quality company bonds, reading below 3.25% as tight and above 5% as wide. It is a corroborating signal: it can agree or disagree with the regime label but does not change it, and it covers US credit, not European.

The yield curve, and an inverted curve

The yield curve is what governments pay to borrow for different lengths of time, plotted from three months out to thirty years. Normally borrowing for longer costs more, so when shorter borrowing costs more instead, the curve is described as inverted, a condition that has historically come before downturns by roughly a year to eighteen months. Sentinel treats an inverted 10-year against 2-year curve as its recession watch signal, which describes conditions today rather than calling a date.