A life insurer's liabilities are long, reasonably predictable in timing and, on guaranteed-savings business, carry a floor under the credited rate. The portfolio's job is to earn more than that floor for as long as the block runs, and the two ways it fails are a duration mismatch and a reinvestment squeeze. The second is the one that kills.
The mismatch runs the other way from the pension's, and that is worth sitting with. A life insurer typically holds assets larger than the reserve, so an asset duration shorter than the liability's does not mean what it appears to mean. Currency duration again settles it: in Worked Example 5 an asset book 1.70 years shorter in duration than its liability is nonetheless hedged to 92.29%, and the duration that would fully hedge is well short of the liability's own. Being over-assets is a partial hedge in exactly the way being under-assets was a leak.
Advertisement
What is the reinvestment squeeze for a life insurer?
The squeeze is the real exposure, and it arrives without any market event. Every rupee that matures must be reinvested at whatever the market offers, and on a guaranteed block the floor does not move with the market. Roll a fifth of a book yielding well above the guarantee into a market yielding below it and the spread narrows immediately; roll another quarter and it narrows again. The worked example takes a comfortable 2.40-point spread down by 74 basis points, almost a third of it, in two tranches with no crash and no default.
Charge the block for maintenance and the exposure resolves into one number a board can act on: the break-even reinvestment rate, which is the guaranteed credited rate plus the expense charge. Everything else is commentary. The distance between that break-even and today's market, multiplied by the share of the book that rolls before the block runs off, is the whole risk.
The guarantee is a written option and the insurer is short it, which is why a fall in rates lands twice. The liability's present value rises faster than the assets', and at the same instant every rupee of new money earns less against a floor that has not moved. Only the first hit appears in a duration table, and that asymmetry is what negative convexity means on the liability side of an insurance balance sheet.
Indian insurers work inside a quantitative investment pattern rather than a prudent-person standard: floors on central government securities and on government and other approved securities, a floor on housing and infrastructure, and a ceiling on other investments with an approved-investment test inside it. (The percentages sit in the IRDAI Investment Regulations and the master circulars under them, and are amended regularly. Check the current text before applying any figure.) The consequence for the squeeze is direct. An insurer who cannot reach for credit or duration cannot outrun a reinvestment problem, so the honest responses are repricing new business, hedging where permitted, or accepting a smaller block.
Insurers, part two: property and casualty, the cycle, and a tax rule that reverses an answer
A property and casualty insurer holds two liabilities that behave nothing alike, and the portfolio has to serve both.
Short-tail business, motor damage and homeowners, pays out inside about a year. Long-tail business, general liability and workers' compensation, pays out over six or seven. Short-tail reserves must be matched with liquid short assets, because a hailstorm settles in months and a portfolio forced to sell long bonds to pay claims will do it badly. Long-tail reserves can support duration and, at the margin, illiquidity. A blended payout duration hides both facts, which is why casualty asset-liability work is done line of business by line of business rather than on the blend.
The underwriting cycle is the second liability the portfolio serves. Investment income and underwriting result draw on one pool of surplus, so the number that sets the portfolio's risk budget is the combined ratio at which investment income exactly offsets the underwriting loss. Worked Example 6 puts it at 107.525000% for a book with $270.9 million of investment income against $3,600 million of net earned premium. An insurer running comfortably below that can take portfolio risk. One running above it is already funding underwriting out of investment income and has no capacity to lose money twice in the same year, and the cycle moves that capacity without asking.
How does the US tax rule change a bond choice?
The tax overlay is where casualty insurers stop resembling any other investor. Under the US proration rule an insurer must reduce its loss-reserve deduction by 25% of tax-exempt interest received (the statute sets that percentage at 5.25% divided by the top corporate rate, which is 25% at today's 21%), so the effective federal tax on municipal interest is 25% of 21%, or 5.250000%, rather than zero. (The rule sits at Internal Revenue Code section 832(b)(5)(B); verify the current proration percentage and the corporate rate before use.)
That is enough to reverse an answer. Worked Example 6 runs a 4.30% municipal against a 5.30% corporate: for the insurer the corporate wins by 11.2750 basis points, and for an ordinary taxpayer at the same 21% rate the municipal wins by 11.3000. Same two bonds, same yields, same headline rate, opposite conclusions, on a rule that lives in one subsection of the code.
One more state, and it is the one people forget. An insurer with loss carry-forwards is not paying tax at all, so the exemption is worth nothing and the municipal simply loses by its full yield disadvantage. A tax-driven allocation held by a non-taxpayer is a straight give-up, and casualty insurers cross in and out of that state with the cycle.
{
"skill": "Reading an institution's stated mandate for the constraint that actually binds",
"use": "Have an assistant extract every quantitative limit from a long investment policy statement, pattern of investment or regulatory schedule into one table, then have it test a proposed policy portfolio against every row at once and report which rows are slack, which are tight and which are breached.",
"tools": ["Document-grounded AI running on the policy document you already hold", "Chat assistants with code execution for the constraint test", "The regulator's own site for the notification or circular the limits come from"],
"prompt": "From the attached investment policy document, extract every quantitative limit into a table with columns: asset class or category as the document names it, minimum, maximum, the basis the percentage is taken on (total assets, incremental flows, a specific fund), the clause reference, and the effective date. List separately every limit that is stated qualitatively rather than numerically. Then take this proposed portfolio [paste weights] and report, row by row, the slack in percentage points against each minimum and maximum, flagging any breach. Do not infer a limit that is not stated; write MISSING where the document is silent.",
"verify": "The extraction is the failure point, and it fails silently. Read every extracted row back against the source clause yourself, because a document model that misreads 'not less than' as 'not more than' produces a table that is internally consistent and exactly backwards. Check the basis on every percentage: a limit on incremental annual flows and a limit on the total corpus are different constraints and are frequently written in adjacent clauses. Date-check the document against the regulator's current notification, since patterns of investment and capital rules are amended by circular and a superseded schedule reads exactly like a current one. Recompute at least two constraint tests by hand. Where the answer will inform a real allocation, cite the clause and the date in the memo so the next reader can check what you checked.",
"diy": "You must be able to read the mandate yourself and say which single constraint is binding and what it costs, because that judgement is the deliverable and no extraction produces it. The table tells you a cap is at 15%; only you can say that the cap is worth 68 basis points a year against what the trustees would otherwise hold, and that the duration decision everyone is arguing about is worth five.",
"market": "IN"
}