Two answers to one survival question
By now you have met the parts: every danger sorted into ERM's boxes, each one measured with Value at Risk or its tail-aware cousins, all read off the one balance sheet where reserves sit as honest valuations of future promises, not idle cash. This guide answers the question all of that was building toward — the one a policyholder would ask if they could: beyond what you already owe me, how much spare money do you keep on hand so that, even after a genuinely bad year, you can still pay? That spare money is capital, and there are two voices that answer how much of it you need.
The first voice is the firm's own. Shut the door, send the regulator home, and ask honestly from your own view of your risks: how much cushion do I truly need to survive a defined bad outcome? That answer is economic capital — the capital the business decides it really needs, computed from its own internal model of all its risks combined. The second voice is the regulator's. It will not take your word for it, so it writes down a rule that every insurer must obey: risk-based capital (RBC), a minimum that scales up with how risky your particular balance sheet is. One number is your truth; the other is the law. A well-run firm holds enough to satisfy both — and watches the gap between them like a hawk.
Economic capital: capital to survive a chosen disaster
Economic capital is built directly on the risk measures from the previous guides. Pick a horizon, almost always one year, and a confidence level — 99.5% is the common European standard, some firms push to 99.95%. Then simulate the change in your net worth across thousands of scenarios that shock markets, defaults, claims and operations all at once. Line up the results from best to worst. Economic capital is the loss at your chosen point in that distribution: with a 99.5% VaR measure, it is the loss you would exceed only one year in two hundred. Hold that much cushion and, by your own model, you survive the 1-in-200 year.
Notice the honesty hard-wired into the definition. 'Survive a defined bad outcome' is a choice, not a fact of nature — pick 99.5% and you have explicitly accepted a 0.5% chance of failing anyway. No amount of capital buys certainty; economic capital simply makes the confidence level you are buying explicit and consistent. And because it is built on a measure, the choice of measure matters: many actuaries set economic capital from a tail average (Tail VaR or expected shortfall) rather than a plain quantile, precisely because VaR is blind to how deep the tail goes. The same firm can report two different economic-capital numbers depending only on which risk measure it picked.
Economic capital is the firm's own truth, and that gives it three jobs no regulatory rule can do as well. It is the benchmark against which the regulator's number is judged adequate or merely passable. It is the engine of pricing — each product can be charged for the capital it consumes. And it lets management compare unlike businesses on one ruler: a motor book and an annuity book earning the same profit are not equally good if one ties up twice the capital. But never forget what it is: a model output, only as trustworthy as its weakest assumption — above all the correlations and the shape of the tail, where data is thinnest. It is an estimate dressed up as a number.
Risk-based capital: the regulator's trip-wire
A regulator cannot let each insurer mark its own homework. So instead of a single flat capital amount for everyone — which would over-burden a safe firm and under-protect a reckless one — it writes a formula that scales the requirement up with the riskiness of the actual balance sheet. In the United States this is risk-based capital, designed by the NAIC. The recipe walks through each kind of risk a company carries — asset default, insurance and mortality risk, interest-rate risk, business risk — and assigns a charge to each by multiplying the exposure by a prescribed factor. A junk bond attracts a bigger asset charge than a government bond; a large book of term insurance attracts a mortality charge. The riskier the cargo, the stronger the brakes the rule demands.
The charges are not simply added — that is the subtle, important part, and we will dwell on it next. They are combined with a square-root 'covariance' formula that grants partial credit for the fact that the risks rarely all go wrong together. The output is the required RBC. The regulator then compares the firm's actual capital to it as a ratio, and falling through tiered thresholds triggers escalating intervention — at first a required plan, eventually the authority to seize the company. Europe's Solvency Capital Requirement plays the same role with a different accent: rather than fixed factors, it is calibrated as a one-year 99.5% VaR, and can be computed with the standard formula or an approved internal model. Both are cousins of economic capital; the difference is one of spirit.
The diversification benefit: why the whole needs less than the sum
Here is the single most consequential idea in capital. Suppose three business lines, looked at alone, would each need 60, 50 and 20 of capital — 130 in total. A firm that simply summed them would be assuming all three suffer their worst year on the very same day. But a motor crash wave, a stock-market slump and a longevity surprise are not the same event; they are imperfectly correlated, so their bad years rarely coincide. When you put them on one balance sheet and run the combined distribution, the requirement might come out at only 100. That 30 gap — the difference between the sum of standalone capitals and the capital the whole truly needs — is the diversification benefit. It is pooling working at the level of the entire enterprise, the same logic that made insurance possible, now applied to risks rather than policyholders.
Where does the square-root formula come from? For two risks measured by their standard deviations, the combined standard deviation is not the sum but the square root of (a-squared + b-squared + 2-rho-a-b), where rho is the correlation between them. The arithmetic below shows the gift correlation gives you. The key takeaway is the direction: the lower the correlation, the larger the saving; at correlation 1 (everything moves together) the benefit vanishes and the capitals simply add. This is exactly why VaR is a worry here — it is not always sub-additive, so it can occasionally report a combined figure larger than the sum and wrongly punish diversification, whereas a coherent measure like Tail VaR never does.
Two risks, standalone capitals 60 and 80. Combined = sqrt(60^2 + 80^2 + 2*rho*60*80) rho = +1.0 -> sqrt(10000+9600...) = 140 (no benefit; just the sum) rho = +0.5 -> sqrt(3600+6400+4800) = 122 diversification benefit 18 rho = 0.0 -> sqrt(3600+6400+0) = 100 benefit 40 rho = -0.5 -> sqrt(3600+6400-4800) = 72 benefit 68 Lower correlation -> bigger saving. In a crisis rho jumps toward +1.
Allocating capital back, and paying its rent
The diversification benefit creates an awkward question, like flatmates who share rent splitting the saving fairly. The whole needs 100, but each line caused some of that need — so how much should each one be charged? This is capital allocation: taking the combined figure and assigning a share back to every business unit, so each product can be measured for the return it earns on the capital it truly consumes. The cleanest method is marginal (Euler) allocation — charge each line for how much it adds to the total when you switch it on. A line that diversifies well against the rest gets a gentle charge; a line that piles onto an existing concentration gets a heavy one. Honest caveat: allocation is not unique. Different reasonable methods give different answers, so it is partly an art with real consequences — it decides which products look profitable and which get repriced or dropped.
Allocation answers 'who consumes the capital?' One question remains: capital is not free, so what does holding it cost, and who pays? Shareholders who lock up capital for years behind uncertain promises demand a return for doing so. That required return, expressed as a rate, is the cost of capital — and it is the bridge from capital to price. Every product's premium must, on top of expected claims and expenses, earn enough to pay the cost of the capital it ties up; a line that consumes a lot of allocated capital must charge more to clear the same hurdle. This is how the abstract capital number reaches all the way down into the price a customer actually pays.
The same cost of capital reappears in a striking place: on the liability side of the balance sheet itself, as the risk margin. Imagine selling your run-off book to a willing buyer. They will not take it for the best-estimate liability alone, because they must lock up capital for years to be sure they can pay even if things go badly — so they demand a little extra for tying that capital up. That extra is the risk margin, computed by the cost-of-capital method: project the regulatory capital needed in every future year until the obligations run off, charge each year a prescribed cost-of-capital rate (historically 6% per year under Solvency II), and discount the rents back to today. The technical provision is then best estimate plus risk margin. The risk margin is not a hidden profit cushion or a slush fund — it is the explicit, honest price of the capital that must stand behind uncertain promises.
Putting it together — and staying honest
Step back and the whole machine comes into view. The firm models its risks together to find economic capital, its own honest cushion. The regulator demands risk-based capital, a comparable floor. Both credit a diversification benefit because risks are imperfectly correlated, and both fall apart if those correlations were optimistic. The combined capital is allocated back to business units so each can be priced to recover its cost of capital, and that same cost-of-capital idea reappears as the risk margin on the liabilities. Finally, comparing the capital a firm actually holds against the capital it needs is exactly the test of capital adequacy — the coverage ratio that boards, rating agencies and regulators all watch.
Now the warnings that keep a good capital analyst humble. First, every one of these numbers is a model output, and the parts you can see least clearly — the tail and the correlations — are precisely the parts that drive the answer. Second, 'adequate' is always relative to a chosen confidence level: a 1-in-200 standard still admits a 1-in-200 chance of failure, and no regime sells certainty. Third, capital adequacy is necessary but not sufficient. A firm can pass every capital test and still fail through a liquidity squeeze — solvent on paper but unable to turn assets into cash fast enough to pay claims when they fall due — or through governance failures the numbers never revealed. Capital answers 'is there enough net worth?', not 'can we pay this Friday?'.
This is where actuarial science meets the boardroom and the regulator. The same disciplined idea that began rungs ago — measure a risk, hold enough to survive it — has grown up into the language a board uses to decide which businesses to grow, what to charge, and how much to reinsure, and the language a regulator uses to protect the public's trust. The numbers are only as good as their assumptions, but the discipline of demanding them, stating them honestly, and acting on them is precisely what ERM is for.