PASS exact identity: with uniform costs and two equal projects the expected least cost is n/(4(n+1)) for every even n up to 80 (exact rational arithmetic)
PASS limiting distribution: realized cost C_n => Z^2/4 (median 0.114, 99th percentile 1.659, ratio of the 99th percentile to the mean 6.63), D/sqrt(n) => |Z|/2 and sqrt(n) price => |Z| with uniform costs: n=2000: mean 0.259, median 0.121, q99 1.558, E[D]/sqrt(n) 0.407, sqrt(n)E[price] 0.791; n=20000: mean 0.250, median 0.118, q99 1.559, E[D]/sqrt(n) 0.401, sqrt(n)E[price] 0.795; n=200000: mean 0.250, median 0.109, q99 1.651, E[D]/sqrt(n) 0.394, sqrt(n)E[price] 0.781
PASS retained veto: the shortfall divided by sqrt(n) behaves like |Z|/2 with the premium retained and |Z|/4 with it returned in equal shares, and the probability of no shortfall falls toward zero: retained, n=4000: mean 0.398, median 0.332, q95 1.044, P(Q=0) 0.039; retained, n=40000: mean 0.379, median 0.323, q95 0.976, P(Q=0) 0.014; returned, n=4000: mean 0.205, median 0.174, q95 0.506, P(Q=0) 0.062; returned, n=40000: mean 0.196, median 0.160, q95 0.510, P(Q=0) 0.028
PASS retained veto, other densities: with g_c(0) = 2 and g_v(0) = 3 the shortfall scales with gamma/2 = g_v/(2 g_c) = 0.75: mean 0.576 against 0.598, 90th percentile 1.231 against 1.234
PASS retained veto: with replacement capital a sqrt(n) the probability of closing tends to P(|Z| <= 2a), a confidence level below one: a=0.5: 0.675, 0.693 (limit 0.683), a=1.0: 0.949, 0.957 (limit 0.954), a=1.5: 0.993, 1.000 (limit 0.997)
PASS binding enrollment, premium retained: sqrt(n) times the enrollment threshold v*_n, the expected burden computed exactly, tends to 1/(sqrt(2 pi) g_c(0)) = 0.3989: n=4000: 0.3950, n=40000: 0.3977, n=400000: 0.3985, n=4000000: 0.3988
PASS binding enrollment, premium retained: the number declining divided by sqrt(n) concentrates at a* = gamma/sqrt(2 pi) (mean 0.398, standard deviation 0.0447 at n = 40,000)
PASS binding enrollment, premium returned: every investor's expected net burden is E[C_n]/n = 1/(4(n+1)); at critical load the probability of closing is (1 - 1/(4(n+1)))^n, which tends to exp(-mu_v) = exp(-1/4) = 0.7788 (n = 100,000: 0.7788)
PASS binding enrollment, premium returned: in every draw the net burdens sum to the realized cost of moves, so by exchangeability each investor's expected net burden is E[C_n]/n (simulated 0.000123 against 0.000125)
PASS premium returned: in 61500 misreports the gain never exceeds (K/n) times the larger gap between adjacent reports of the others at the margin (largest ratio 0.500, bound 0.5)
PASS retained veto: the investor whose report sets the price gains in expectation (2.07e-02) by shading it toward the next report, because a lower price raises the probability that the book closes
PASS correlated example beyond counts: the scheme of liability 6.5 makes formation the unique correlated equilibrium for risk-neutral investors (largest failure probability -0.0e+00) but admits a correlated equilibrium that fails for certain under constant absolute risk aversion 0.05 (1.000)
PASS correlated robustness under constant absolute risk aversion: minimizing liability over utility certificates never goes below R* for counts or R_c in general (36 structures; smallest margin 0.109)
PASS band of discretion: with posterior 0.9 on A, v+ = 2 and v- = -1, a CARA investor with coefficient 1 values A at 0.932 and A+B at 1, so the less favored project can raise a risk-averse certainty equivalent although its expected value is -0.7
PASS uncompensated flexibility: with two projects of one unit, investor 1 restricted to A and investor 2 favoring A with a cost 1 of B and a gain 2 from formation, flexibility gives investor 2 a payoff of 1 and a narrow mandate gives 0, so a narrow mandate is not dominant when it prevents closing
PASS interim types: with uniform costs, exact rational arithmetic confirms that the order-statistic recurrence turns the bound (1/2) sum_d [p_d e(d,K+d-1) - p_(d+1) e(d,K+d)] into (1/2) sum_d p_d d/(K+d) [e(d,K+d) + e(d+1,K+d)] for K = 2,...,29
PASS interim types, premium returned: the exact expected net burden of every cost type never exceeds the bound, and n times the bound stays bounded (near 1/2 for uniform costs): uniform, n=400: n*bound 0.480, n*largest burden 0.480; uniform, n=1600: n*bound 0.490, n*largest burden 0.490; uniform, n=6400: n*bound 0.495, n*largest burden 0.495; enrollment-weighted, n=400: n*bound 0.360, n*largest burden 0.360; enrollment-weighted, n=1600: n*bound 0.368, n*largest burden 0.368; enrollment-weighted, n=6400: n*bound 0.371, n*largest burden 0.371
PASS interim types: the exact expected net burden agrees with simulation of the uniform-price assignment at n = 100: c=0.05: exact -0.01232, simulated -0.01239 (se 0.00013); c=1.0: exact 0.00460, simulated 0.00456 (se 0.00019)
PASS interim types: with C = 0.495 and uniform margins, (1 - C/n)^n is 0.6096 at a million investors, against exp(-C) = 0.6096
PASS binding enrollment against a retained veto, premium retained, 99 percent reliability: the exact binding reserve as a share of the limiting veto reserve is 0.427 at n=10000 (55 against 129), 0.376 at n=100000 (153 against 407), 0.346 at n=1000000 (446 against 1288), falling toward the limit 0.310
PASS leave-one-out rebate, the example (0.9, 0.8, 0.1, -0.5): with equal shares the third investor gains 0.30 by reporting 0.7; with the leave-one-out rebate he gains 0.00
PASS leave-one-out rebate: in 73200 misreports across random books of 4 to 8 investors no misreport gains (largest gain 0.0e+00), while equal shares allow gains up to 0.450
PASS leave-one-out rebate: the rebates never exceed the premium, the net burdens sum to C_n + L_n in every draw, and L_n = K(K+1)/n times the gap between the price and the next lower cost on the crowded side, checked against the direct computation
PASS leave-one-out rebate, uniform costs: the exact expected leftover is 0.4461 at n=200, 0.4824 at n=2000, 0.4944 at n=20000, 0.4982 at n=200000, tending to 1/(2 g_c(0)) = 0.5; simulated 0.4925 at n=2000
PASS leave-one-out rebate: the leftover is approximately exponential with mean 1/(2 g_c(0)): uniform costs median 0.331 against 0.347, P(L > 0.5) 0.359 against 0.368; costs with density 2 at zero, mean 0.249 against 0.25
PASS leave-one-out rebate, uniform costs and margins: the probability of closing at critical load is 0.4736 at n=100,000 against exp(-3/4) = 0.4724; two or three extra invitations give 0.9595 and 0.9927
PASS interim types under the leave-one-out rebate, uniform costs: n times the largest expected net burden is 0.960 at n=400, 0.980 at n=1600, 0.990 at n=6400, 0.995 at n=25600; at n=100 simulation gives 0.926 against 0.920
PASS total net burden under the leave-one-out rebate, uniform costs, n=20000: mean 0.760, median 0.605, 99th percentile 2.835, against 0.75, 0.591 and 2.836 for chi-squared(3)/4 (ratio of the 99th percentile to the mean 3.78); Kolmogorov-Smirnov p-value 0.66
PASS retained veto, dominant strategies: in 500 cases (uniform price and leave-one-out rebate, books of 6 to 20 investors) the investor is placed in A exactly when his report exceeds the price he faces, his transfer is constant on each side, and the two transfers differ by that price
PASS retained veto, the bound for one investor: with uniform costs and margins, over 40,100 pairs of the price faced and the fee for the uncrowded project, the expected withdrawal less half the expected transfer collected is never below (1 - p) p / 4 (smallest margin 1.9e-06)
PASS retained veto, mechanisms with dominant strategies: with uniform costs and margins and n=1000, the bound n E[|w|(1-|w|)]/4 = 6.06 on E[Q] - E[revenue]/2, with w the price an investor faces, holds for every member of a family that pays every investor theta times the price he faces less a fee tau0/n, and no member with nonnegative expected revenue loses fewer investors than the bound: theta=0.0, tau0=0.0: E[Q] 12.77, revenue 11.73; theta=0.0, tau0=2.0: E[Q] 13.84, revenue 13.35; theta=0.25, tau0=0.0: E[Q] 9.61, revenue 5.80; theta=0.25, tau0=2.0: E[Q] 9.92, revenue 7.16; theta=0.5, tau0=0.0: E[Q] 6.19, revenue -0.63; theta=0.5, tau0=2.0: E[Q] 6.54, revenue 1.42; theta=0.75, tau0=0.0: E[Q] 2.99, revenue -6.57; theta=0.75, tau0=2.0: E[Q] 3.99, revenue -4.83; theta=1.0, tau0=0.0: E[Q] 0.00, revenue -13.01; theta=1.0, tau0=2.0: E[Q] 1.00, revenue -11.31
PASS retained veto, leave-one-out rebate: E[Q]/sqrt(n) is 0.205 at n=4000 and 0.196 at n=40000, against the bound gamma E|Z|/4 = 0.1995, and the expected transfers collected from those who accept are 0.361 and 0.361, so no fee is needed; with margins six times as dense at zero (gamma = 6, n=20000) they are -0.25 without a fee and 0.74 with tau0 = 1, and E[Q]/sqrt(n) is 1.244 against 1.197
PASS expected-externality transfers: they sum to zero (largest |sum| 3e-16); sqrt(n) pbar_n(0) is 0.3989 against 1/sqrt(2 pi) = 0.3989 and n E[pbar_n(c_1)] is 0.5020 against 1/(2 g_c(0)) = 0.5 at n=40000; with truthful reports under a retained veto the conditional mean of the shortfall, sum_i G_v(b_i+), averages 0.636 (se 0.006) at n=4000, 0.649 (se 0.008) at n=40000, 0.656 (se 0.017) at n=400000 against the limit 0.659; the shortfall itself averages 0.635 (se 0.016) at n=4000 and 0.701 (se 0.023) at n=40000, with P(Q = 0) 0.547 and 0.511 against 0.535 and 99th percentile 3 (limit P(Q <= 3) = 0.991); the unpaid transfers of those who withdraw average 1.4e-04 in absolute value
PASS retained veto, expected-externality transfers: an investor with cost 0.5 gains by reporting a cost of zero when his margin is below the threshold v_n, with sqrt(n) v_n equal to 0.383 at n=4000, 0.394 at n=40000 against 1/sqrt(2 pi) = 0.399; simulated gains: n=4000, margin 0.5 v_n: +1.8e-03 (se 7e-05), margin 0.9 v_n: +4.5e-04 (se 9e-05), margin 1.5 v_n: -1.1e-03 (se 1e-04); n=40000, margin 0.5 v_n: +5.4e-04 (se 4e-05), margin 0.9 v_n: +3.7e-05 (se 5e-05), margin 1.5 v_n: -5.4e-04 (se 6e-05); the expected number of investors with cost at least 0.5 and margin below v_n, divided by sqrt(n), is 0.197 at n=40000 against 0.5/sqrt(2 pi) = 0.199; if all of them report a zero cost, the shortfall at n=40000 averages 19.0 (0.095 sqrt(n)), against 0.70 with truthful reports
PASS the parity device: n=4000: E[Q]/sqrt(n) 0.201, P(Q=0) 0.532, largest Q in an odd state 0, deficit in odd states mean 0.511 (KS p-value against Z^2/2: 0.39), 99th percentile of the deficit 4.30, capacity Q + Delta/2 that suffices with probability 0.4: 0.4, with probability 0.6: 0.118 sqrt(n) (limit 0.127 sqrt(n)), share of states with fewer than 5 percent wrong 0.507; n=40000: E[Q]/sqrt(n) 0.211, P(Q=0) 0.486, largest Q in an odd state 0, deficit in odd states mean 0.470 (KS p-value against Z^2/2: 0.42), 99th percentile of the deficit 3.80, capacity Q + Delta/2 that suffices with probability 0.4: 0.4, with probability 0.6: 0.165 sqrt(n) (limit 0.127 sqrt(n)), share of states with fewer than 5 percent wrong 0.480; the bound gamma E|Z|/4 = 0.1995
PASS the states in which a mechanism is right about all but 5 percent of its investors (n=4000): loo: 0.000, theta(0.5,): 0.000, theta(0.0,): 0.000, theta(1.0,): 0.000, rare(0.5, 0.1): 0.000, parity: 0.496; none exceeds one half
PASS the rare-state mechanism (a charge of 0.5 on every other investor when investor 1 reports a cost below a threshold): n=4000: threshold 0.1010, P(Q=0) 0.897, E[T] +79.87 (se 8.04), E[Q]/sqrt(n) 3.206, median cash deficit 0.294 sqrt(n), (E[Q] + E[Delta]/2)/sqrt(n) 3.385; n=40000: threshold 0.0319, P(Q=0) 0.976, E[T] +157.01 (se 54.33), E[Q]/sqrt(n) 2.364, median cash deficit 0.321 sqrt(n), (E[Q] + E[Delta]/2)/sqrt(n) 2.565; E[Q]/sqrt(n) and the capacity respect the bound 0.1995 although the book almost always closes without replacement capital
PASS the capacity inequality (n=4000, finite-n bound n E[|w|(1-|w|)]/4 = 0.1955 sqrt(n)): (E[Q] + E[Delta]/2)/sqrt(n) is 0.391 (places 0.391, cash 0.000; se 0.011) for theta=0.0, tau0=0.0, 0.419 (places 0.419, cash 0.000; se 0.011) for theta=0.0, tau0=2.0, 0.204 (places 0.199, cash 0.005; se 0.006) for theta=0.5, tau0=0.0, 0.201 (places 0.201, cash 0.000; se 0.006) for theta=0.5, tau0=2.0, 0.199 (places 0.000, cash 0.199; se 0.005) for theta=1.0, tau0=0.0, 0.194 (places 0.016, cash 0.178; se 0.005) for theta=1.0, tau0=2.0, never more than three standard errors below the bound
PASS expected-externality transfers under binding enrollment: in 300 books of 2,000 investors the transfers sum to zero and the net burdens sum to the realized cost of moves in every draw, so each placed investor expects E[C_n]/n and the shortfall is Poisson with mean gamma/4 (mean of the sums 0.270 against E[C_n] = 0.250)
PASS verifiable favored projects: with uniform costs the expected cash paid to the investors moved, E[D X_(D+1)], is n/(2(n+1)) for every even n up to 80, twice the expected cost of the moves (exact rational arithmetic)
PASS verifiable favored projects: in 17343 misreports of cost across books of 4 to 10 investors no misreport gains (largest gain 0.0e+00); at n=20000 no investor withdraws in any of 1,500 books, the cash paid averages 0.476 against 0.5, and a Kolmogorov-Smirnov test against Z^2/2 gives p-value 0.69
41 checks; 0 failures
