PASS compatible capital: the max-flow shortfall equals max_J [K(J) - b(N(J))]+ in 400/400 random books, and widening a permission never raises it (400/400)
PASS compatible capital: two units authorized only for A leave B* = 1; authorizing one of them for B too gives B* = 0
PASS commitment support: widening permissions enlarges the family of forming coalitions and weakly lowers R* (458/458); with capital equal to requirements R* = sum h - max h whatever the assignment of exposure (230/230)
PASS mandate diversification (two projects): the least flexible capital for 99% reliability is z*sqrt(N) asymptotically: N=100: F=22 (0.854), N=1000: F=78 (0.958), N=10000: F=254 (0.986), N=100000: F=811 (0.996)
PASS mandate diversification: N = 10,000 and F = 321 give a Hoeffding bound below 1%; the exact least F is 254
PASS mandate diversification (M projects): flexible capital sqrt(8NM log(M/delta)/3) keeps failure below delta = 5%: N=1998,M=3: F=256, failure 0.000; N=5000,M=5: F=555, failure 0.001; N=20000,M=4: F=967, failure 0.000
PASS systematic mandates (Beta(20,20) mixing): the least flexible share converges to d/(1+d) = 0.2334: N=1000: 0.2380, N=10000: 0.2338, N=100000: 0.2334
PASS common recommendation: if a share alpha = 0.2 of mandates follow one recommendation, the flexible share needed tends to alpha/(1+alpha) = 0.1667 (found 0.1708 at N = 200000)
PASS nominal splitting does not diversify: blocks of 16 investors sharing one mandate raise the flexible requirement by about sqrt(16) = 4 (ratio 3.97)
PASS price of discretion: with idiosyncratic mandates and U[0,1] switching costs the least total cost tends to (M-1)/(4g(0)): M=2 N=1000: 0.248, N=10000: 0.253, N=100000: 0.248; M=3 N=999: 0.518, N=9999: 0.501, N=99999: 0.493
PASS pooled exposure versus a book: at N = 10,000, averaged over 400 draws, pro rata exposure costs 2,500 in misallocation against 0.253 for the efficient book (limits 2,500 and 0.25)
PASS price of discretion: the premium on the over-subscribed project scales like N^(-1/2): sqrt(N)*E[Delta] = 0.754 (N=1000), 0.753 (N=10000), 0.759 (N=100000)
PASS price of discretion: moving the cheapest supporters of each over-subscribed project is optimal among all fractional assignments and never costs more than pro rata exposure (60/60)
PASS systematic mandates carry a price: per-investor cost tends to E[(q-1/2)^2/(2 max(q,1-q))] = 0.00494 (simulated 0.00485 at N=10000, 0.00422 at N=100000)
PASS mechanism: the uniform-price assignment (A slots to the highest value differences at the (K+1)-th bid) admits no profitable misreport in 13200 tests
PASS mechanism: paying switchers rewards an indifferent investor for declaring the crowded project (mean gain 0.167); pricing the crowded project gives no such gain (0.00e+00)
PASS information: the minority project is worth financing exactly when the imbalance lies within a band whose width log((1-p)/p)/log((1+2e)/(1-2e)) falls with informativeness: e=0.001: 127.7, e=0.01: 12.8, e=0.05: 2.5, e=0.1: 1.3
PASS records: the least certainty-equivalent loss from k common exposure profiles is half the sum of the squared singular values of Sigma^(1/2) W* Gamma^(1/2) beyond the k-th (30 random economies)
PASS records: the optimal number of profiles falls with the cost of a record and with the number of constraint classes (k* = 4, 3, 1; with four classes 2)
PASS records: hedging demands proportional to one vector still give a holdings matrix of rank two
PASS flexible share: with a share phi of flexible investors and U[0,1] costs, the program is financed with probability near one and the least cost tends to (m-1)/(4 phi g(0)): phi=0.1: 2.279 (limit 2.500, financed 1.000), phi=0.25: 1.009 (limit 1.000, financed 1.000), phi=0.5: 0.512 (limit 0.500, financed 1.000)
PASS cost bound: with costs whose density vanishes at zero (Beta(2,1)), the least cost stays below E[c] sum_j sqrt(n rho_j(1-rho_j)) and grows like n^(1/4): N=2000: 1.87 (bound 29.8), N=20000: 3.38 (bound 94.3), N=200000: 5.57 (bound 298.1)
PASS fixed cost of moving: with costs at least 0.05 the least cost grows like sqrt(n) while its share per investor vanishes: N=2000: 1.13, N=20000: 3.05, N=200000: 9.89
PASS a pool of order sqrt(n): with 3 sqrt(n) flexible investors the cost of the discretion used grows like sqrt(n): N=10000: 7.7, N=100000: 24.4
PASS two-project VCG with one place left vacant: holders of A pay max(0, w_(K+1)) and holders of B pay max(0, -w_(K)), so only the over-subscribed project is priced and nobody pays when neither is (17944 payments checked)
PASS rebates: returning the premium to the investors moved gives a B-favourer a mean gain of 0.92 from declaring A, while equal-share rebates leave a largest gain of 0.0000
PASS flexibility under the uniform price: reporting a finite cost truthfully is never worse than a narrow mandate (2000 tests)
PASS willingness: enlarging the set of willing investors weakly lowers R*(Gamma, W) (182/182)
28 checks; 0 failures
