Let be the discount level during policy year , with . Under the usual assumption of independent successive annual accidents and repair costs with the same laws each year, the submitted-claim decision depends only on and that year's cost. Consequently the conditional distribution of depends on the past only through : this is the Markov property. The no claims discount system is therefore a time-homogeneous Discrete-time Markov chain. Stationarity of accident probabilities alone, without this independence assumption, would not by itself establish the Markov property.
At level one a claim keeps the policyholder at one, while no submitted claim moves them to two. At level two, a submitted claim moves them to one and no claim to three. At level three, a claim moves them to two and no claim keeps them at three. Thus in the specified transition matrix is the annual probability of a submitted claim at current level , not merely of an accident. A small accident paid privately counts as no claim for these transitions.
Let denote the undiscounted annual premium. It is not specified numerically in the source; costs can instead be measured in units of that premium, giving . The actual premium payments at the three levels are , , and . Compare the two future paths, assuming no further accidents, to obtain the two-year premium-loss claim threshold. From level one, a current claim leads to future levels , whereas no claim leads to , giving extra premium
From level two the two paths are and , and from level three they are and . Hence
These are all strictly positive. The PDF's discount fractions are and ; the converted TeX has lost these factors in the percentages.
For a repair cost with the specified lognormal distribution, . Put
where is the standard normal distribution function. Conditional on an accident, the submitted-claim probability is . Multiplying by the accident probability gives
Set for the premium-unit convention. Without a monetary premium scale, discount percentages alone cannot determine a numerical threshold for a monetary repair cost.
This nearest-neighbour Markov chain is a finite birth-death chain. Its stationary distribution satisfies the endpoint balance equations
These detailed balance equations also imply the middle-state balance equation. The three-level no claims discount equilibrium is therefore
where
Since each lies strictly between zero and one, this is an irreducible Markov chain. Positive holding probabilities at the endpoints also make it an aperiodic Markov chain. Its unique stationary distribution is the limiting distribution even from the initial level one, and gives the steady-state proportions for a homogeneous population of such policyholders.
A submitted claim at level one conditions the repair cost on . The accident probability cancels in this conditional expectation. For , the truncated lognormal moment follows by completing the square:
Dividing by yields the expected size of an actual submitted claim:
The unconditional expected amount paid per year from level one would instead multiply its tail first moment by ; it is not the conditional claim size requested here.
The surplus in the classical risk model is . With ruin defined as a strictly negative surplus, its ultimate ruin probability is
Let . The first arrival time has exponential distribution of rate , independent of the first claim. Before that claim no ruin occurs, because the premium rate is positive. Given its arrival at and size , available capital just before it is . If ruin is immediate; otherwise the independent increments and fresh claim sizes make subsequent ruin probability . Thus first-claim conditioning gives
This is the ruin counterpart of the first-claim decomposition for survival probability. Define
The convolution term is continuous: extend the bounded function by zero to negative arguments and use continuity of translations in for the integrable density . The tail function is also continuous. Changing variables to yields
Differentiating this exponentially weighted integral establishes differentiability rather than presuming it. The fundamental theorem of calculus gives
Reversing the integration variable in the convolution therefore proves the ruin integro-differential equation:
At zero the derivative is the right derivative; the same representation supplies it.
For the exponential claim law, introduce
Differentiating the explicit convolution gives . Hence
using the proved integral equation. Differentiating that equation once more gives
so the required ordinary differential equation is
Positive relative safety loading means , so . The general solution is .
The condition at large capital must also be used. The strong law of large numbers for the Poisson process and the independent claims gives almost surely. Thus . Its path has only finitely many jumps on each bounded time interval, so its supremum over all time is finite almost surely. Therefore as , forcing . The supplied initial value then gives ultimate ruin with exponential claims:
The decay rate is also the exponential-claim adjustment coefficient.
For the maximum aggregate loss in a classical risk model, the ruin event is exactly . Since , , and its full cumulative distribution function is
In particular it has an atom of mass at zero. Conditional on , its distribution is exponential with rate ; omitting that atom would give an incorrectly normalized law.
The claim law is an Erlang distribution of shape two and rate . Direct integration, using for , gives its moment-generating function
Differentiating at zero gives . The positive relative safety loading condition for the classical risk model is therefore
The adjustment coefficient is the positive root in the finite-transform domain of
Set and . For , division by reduces the equation to
The two roots are . The minus root lies strictly between zero and one: positivity follows from , and the numerator is smaller than . The plus root exceeds one and is outside the moment-generating function's domain. Consequently the adjustment coefficient for shape-two Erlang claims is
The zero root of the original equation is not an adjustment coefficient. The increasing secant slope on also establishes uniqueness of the positive root.
Now set . Under quota share reinsurance, the retained payment is . For , the change of variables yields
Here the second gamma parameter is the rate. In particular, and . At full cession the retained payment instead has a point mass at zero.
The reinsurer's expected annual payments are , so its loaded premium is . Subtract it from the original premium income to obtain the direct insurer's net premium rate
Its expected retained claim outflow is . Thus
by the stipulated lower bound on . In particular , so the retained portfolio satisfies the net profitability condition. Its relative safety loading is .
Apply the same adjustment coefficient for shape-two Erlang claims formula with scale and net premium . Writing
gives
Equivalently it is the positive solution, with , of
For fixed positive initial capital, a larger adjustment coefficient reduces the exponential upper bound from the Lundberg inequality. It also makes the large-capital decay exponent in the Cramér–Lundberg ruin asymptotic larger. These are the reasons for maximizing as a ruin-risk criterion; the coefficient alone need not order exact finite-capital ruin probabilities for arbitrary different models.
If , then and is constant. The equal-loading quota share and vanishing retention relation becomes
The numerator is positive for , so decreases strictly with and tends to infinity as . There is no maximizing retention in the stipulated interval ; the best limit is full cession. If the endpoint is admitted as an idealized contract, all claims are transferred and the net premium income is zero, leaving the direct insurer's capital constant and its ultimate ruin probability zero. At that endpoint the coefficient equation is identically zero, so it no longer defines a unique finite adjustment coefficient. Calling optimal requires this explicit extension of the original admissible set.
Assume , so the residual sum of squares is positive almost surely. Write and . The log-likelihood is
The decomposition shows that, at every fixed positive variance, its constrained maximum-likelihood estimate of the mean is . Substituting this mean and differentiating with respect to the variance gives
The denominator is required by maximum likelihood estimation; is the unbiased variance estimate, which is a different estimator.
Under the null variance , maximize over the same mean . Subtracting the null maximum log-likelihood from the unrestricted maximum gives the likelihood-ratio test statistic
By Cochran's theorem, the residual sum of squares in a normal sample satisfies under the null. Set , so . Then
The normal mean is a nuisance parameter, but this null law does not depend on it.
A Bartlett correction removes the first-order discrepancy between the null mean of a regular likelihood-ratio test statistic and the degrees of freedom of its limiting chi-squared distribution. In the convention used here, if the limiting degrees of freedom are and
then is the Bartlett correction coefficient, also called the Bartlett correction factor, and the Bartlett-corrected likelihood-ratio statistic is . Its null mean is . The finite- divisor is , rather than itself. There is one restriction on the variance here, so .
Expand the logarithm around and use the permitted termwise asymptotic integration:
Using the supplied raw moments of the chi-squared distribution and expanding gives
For instance, and give the first formula after subtracting . The same binomial expansion, using the raw third and fourth moments, gives the other two formulas.
We must justify why higher moments cannot change the coefficient. Represent for independent standard normal variables, and put . Each is centered and has all absolute moments finite. With , one has . The supplied centered-sum bounds say, for each fixed integer ,
Also and . Therefore, for every fixed , expanding gives
These are bounds on the absolute values of signed moments; the odd-moment estimate must not be misread as a bound on . In particular, each fixed Taylor term of degree has or smaller. Combined with the question's permission to integrate the asymptotic expansion term by term, this controls the higher-order contribution at the required order. It does not require a globally convergent Taylor series for every realized .
Substituting the first three moment formulas now gives
Thus the Bartlett correction for a normal variance with unknown mean is
Its null mean is , and the usual reference law is . Keeping the estimated mean's loss of one residual degree of freedom, and using the maximum-likelihood variance divisor , are both essential to the coefficient .
For a regular two-parameter model, write for the parameter of interest and for the nuisance parameter. If are its score functions, the parameters are orthogonal statistical parameters when the cross entry of the expected Fisher information matrix vanishes:
This condition may hold at a particular parameter value or throughout the parameter domain. It concerns the information geometry, not exact independence of finite-sample estimators. Under regular asymptotic efficiency, a diagonal information matrix gives zero first-order asymptotic covariance.
An interest-respecting reparametrization leaves unchanged, or replaces it by a one-to-one function of alone, while changing the nuisance coordinate by a smooth invertible transformation which may depend on both old parameters. Thus it preserves the level sets of the inferential target. Without loss of generality, keep itself and write the old nuisance as , where is the new nuisance coordinate and .
The chain rule for the log-likelihood gives new score functions
Their expected cross product is
Setting this to zero, with , gives the orthogonalization equation for two statistical parameters
Solve this ordinary differential equation along curves of constant , using to label distinct initial conditions. Smooth local solutions with nonzero transverse derivative produce an invertible interest-respecting reparametrization. This informal construction identifies precisely which derivative is held at fixed new nuisance coordinate.
For the given positive-shape generalized Pareto distribution, its one-observation log-likelihood is
Define and . The density transformation , , gives
Thus has an exponential distribution of mean . Direct differentiation of the original log-likelihood, holding the other original parameter fixed, gives
The following exponential integrals determine their expected products:
For example, follows by integrating the exponential density, and differentiating this integral in gives . These formulas also give .
For clarity, the nuisance information and cross information can be calculated explicitly. Put . Then
Using , the cross product is
The same moments give . Therefore the Fisher information matrix for one observation is
In particular, for every admissible pair, proving the original shape and scale parameters are not orthogonal. No finite-mean assumption on was needed: the transformed logarithm and the bounded variable have the required integrable score products even when the Pareto mean is infinite.
The orthogonalization equation for two statistical parameters now becomes
Integration along a constant- curve gives . Hence a global interest-respecting reparametrization is
It is smoothly invertible on the entire positive parameter domain, with inverse . The Fisher information matrix transforms as , where
Substitution gives
This verifies the orthogonal shape and scale for a generalized Pareto distribution directly, as well as by the differential equation. For an iid sample, every information entry is multiplied by the sample size, so the same parameter orthogonality remains valid.
Write for the cumulant-generating function in the question's parametrization. Multiplying the density by replaces in the decaying part of its exponent by . For , the remaining integral is the normalizing integral for an Inverse Gaussian distribution with parameters . Comparing the two normalizing constants gives
In particular , so the mean parameter used in the more usual Inverse Gaussian distribution notation is ; it must not be confused with the present parameter .
First compute the cumulant-generating function of the sample average, as requested. Independence gives
This is exactly the cumulant-generating function of an Inverse Gaussian distribution with parameters . The uniqueness theorem for moment-generating functions therefore identifies that law. Scaling and including the density Jacobian gives
Equivalently, the inverse Gaussian sum closure is in the parametrization of this question. Its mean is and its variance is , in agreement with addition of independent means and variances.
For a general continuous iid sample with cumulant-generating function , finite near zero, set and choose a real saddle in the interior of the finite moment domain satisfying . With , the saddlepoint density approximation for the sum is
Under the usual smooth nonlattice conditions, at a fixed interior the true density has this leading expression times . This is a relative density approximation; the leading expression is not automatically an exactly normalized probability density in every model.
One way to understand the statistical saddlepoint approximation is through exponential tilting. The tilted one-observation density is , with mean and variance . Undoing this tilt for the sum multiplies its tilted density at by . At , is exactly the tilted mean of the sum. A local version of the central limit theorem gives tilted density there approximately . Multiplying these two factors gives the displayed saddlepoint density approximation. Higher local expansion terms produce the stated relative error order.
For the Inverse Gaussian distribution, differentiation gives
Thus for every the saddle exists, with
At this saddle,
Substitution into the saddlepoint density approximation gives precisely the exact density already found for . Therefore
This exact saddlepoint density of an inverse Gaussian sum is stronger than a relative claim: the leading formula is exact and needs no renormalization.
For the final distribution, both tails decay polynomially. Any positive exponential tilt makes the positive tail nonintegrable, and any negative tilt makes the negative tail nonintegrable. Its moment-generating function is therefore infinite away from zero, leaving no open interval on which to form the cumulant-generating function and solve a saddle equation. Here polynomial tails prevent exponential tilting, so the conventional saddlepoint density approximation is unavailable. A finite mean and variance, and even an ordinary central limit theorem, do not supply the missing exponential moments.
A useful full statement is the rank-sum form of Cochran's theorem. Let and let be real symmetric matrices satisfying . Put and . Then , and the following conditions are equivalent: the ranks sum to ; the are orthogonal projection matrices with mutually orthogonal ranges; and the are independent, with . A rank-zero form is identically zero. In particular, the usual projection version of Cochran's theorem partitions the squared length of a standard normal vector into independent chi-squared distributions.
Here is a proof which also explains the rank condition. Let . Since every vector satisfies , the spaces span . Therefore their dimensions sum to at least . If they sum to , their sum is a direct sum, so every vector has a unique decomposition into vectors belonging to the . For , the identity is one such decomposition, while the decomposition with just in position and zero elsewhere is another. Uniqueness gives . Thus
A symmetric matrix with this idempotence is an orthogonal projection matrix. Its different ranges are orthogonal: for , , their inner product is . Choose an orthonormal basis for each range and concatenate them. The resulting change-of-basis matrix is an orthogonal matrix. In these coordinates, each is the sum of squares of the coordinates belonging to its range.
The transformed multivariate normal distribution still has density proportional to , which factors into the standard normal densities of its coordinates. The coordinates are independent, so the disjoint groups of squared coordinates give independent chi-squared distributions of the asserted degrees of freedom. This proves the forward implication. Conversely, if the have the asserted chi-squared laws, then , while has expectation . Taking expectations yields . This completes the equivalence and the proof of Cochran's theorem.
For the application, specify a normal linear model
with fixed full-column-rank design matrix of size , , unrestricted , and unknown . Test against , where has independent rows, , and is specified. The null contains an affine space of means of dimension , whereas the full mean space has dimension .
Let be the ordinary least squares and maximum-likelihood estimate in the full model, and write . The constrained estimate follows by minimizing the residual squared length under :
Indeed, and the correction is the required constrained projection in the inner product. The least-squares normal equations give the Pythagorean decomposition
Consequently the explicit residual quantities in the likelihood-ratio test statistic are
Here is the full-model residual sum of squares, and is the increase caused by imposing the null restrictions.
To see the geometry behind Cochran's theorem, choose any with , and let project onto and onto . For ,
The projectors and are orthogonal and have ranks and . Under , the mean of belongs to , so both of these projectors annihilate it. Thus Cochran's theorem gives
The nuisance variance cancels in the nested-model F-test statistic
Since the likelihood-ratio test statistic is
it is a strictly increasing function of . Therefore an exact level- likelihood-ratio test rejects when , where the right-hand side is the upper reference quantile of the F-distribution. This calibration is exact under the normal linear model, rather than an asymptotic chi-squared replacement.

Pinned article: Introduction to the OurBigBook Project

Welcome to the OurBigBook Project! Our goal is to create the perfect publishing platform for STEM subjects, and get university-level students to write the best free STEM tutorials ever.
Everyone is welcome to create an account and play with the site: ourbigbook.com/go/register. We belive that students themselves can write amazing tutorials, but teachers are welcome too. You can write about anything you want, it doesn't have to be STEM or even educational. Silly test content is very welcome and you won't be penalized in any way. Just keep it legal!
We have two killer features:
  1. topics: topics group articles by different users with the same title, e.g. here is the topic for the "Fundamental Theorem of Calculus" ourbigbook.com/go/topic/fundamental-theorem-of-calculus
    Articles of different users are sorted by upvote within each article page. This feature is a bit like:
    • a Wikipedia where each user can have their own version of each article
    • a Q&A website like Stack Overflow, where multiple people can give their views on a given topic, and the best ones are sorted by upvote. Except you don't need to wait for someone to ask first, and any topic goes, no matter how narrow or broad
    This feature makes it possible for readers to find better explanations of any topic created by other writers. And it allows writers to create an explanation in a place that readers might actually find it.
    Figure 1.
    Screenshot of the "Derivative" topic page
    . View it live at: ourbigbook.com/go/topic/derivative
  2. local editing: you can store all your personal knowledge base content locally in a plaintext markup format that can be edited locally and published either:
    This way you can be sure that even if OurBigBook.com were to go down one day (which we have no plans to do as it is quite cheap to host!), your content will still be perfectly readable as a static site.
    Figure 5. . You can also edit articles on the Web editor without installing anything locally.
    Video 3.
    Edit locally and publish demo
    . Source. This shows editing OurBigBook Markup and publishing it using the Visual Studio Code extension.
  3. https://raw.githubusercontent.com/ourbigbook/ourbigbook-media/master/feature/x/hilbert-space-arrow.png
  4. Infinitely deep tables of contents:
    Figure 6.
    Dynamic article tree with infinitely deep table of contents
    .
    Descendant pages can also show up as toplevel e.g.: ourbigbook.com/cirosantilli/chordate-subclade
All our software is open source and hosted at: github.com/ourbigbook/ourbigbook
Further documentation can be found at: docs.ourbigbook.com
Feel free to reach our to us for any help or suggestions: docs.ourbigbook.com/#contact