# Cultural distance from China to five Southeast Asian markets (and the United States, for contrast),
# with the Kogut & Singh (1988) index on the six Hofstede dimensions as published in Hofstede,
# Hofstede & Minkov (2010): power distance, individualism, masculinity, uncertainty avoidance,
# long-term orientation, indulgence.
#   KS_j = (1/n) * sum_i (I_ij - I_i,China)^2 / Var_i
# Var_i should be the variance across all countries in the database; here it is the variance across
# these seven countries only, so the index is for illustration. (Later online versions of the
# scores show Malaysia's power distance capped at 100 instead of 104.) An entry-friction score then adds
# two non-cultural layers whose weights are the author's assumptions, to show how rankings move.
from statistics import pvariance

DIMS = ("PDI", "IDV", "MAS", "UAI", "LTO", "IVR")
H = {"China":     (80, 20, 66, 30, 87, 24),
     "Malaysia":  (104, 26, 50, 36, 41, 57),
     "Indonesia": (78, 14, 46, 48, 62, 38),
     "Thailand":  (64, 20, 34, 64, 32, 45),
     "Vietnam":   (70, 20, 40, 30, 57, 35),
     "Singapore": (74, 20, 48, 8, 72, 46),
     "USA":       (40, 91, 62, 46, 26, 68)}
VAR = [pvariance([v[i] for v in H.values()]) for i in range(6)]

def ks(c, dims=range(6)):
    return sum((H[c][i] - H["China"][i]) ** 2 / VAR[i] for i in dims) / len(dims)

# Non-cultural layers, 0 (none) to 1 (high), as entered by an analyst for a consumer-goods entrant:
#   halal: share of the mass market where halal certification is expected or required
#   rules: foreign-ownership limits and licensing friction for the planned business (assumed)
LAYERS = {"Malaysia": (0.7, 0.4), "Indonesia": (0.9, 0.7), "Thailand": (0.1, 0.6),
          "Vietnam": (0.0, 0.5), "Singapore": (0.2, 0.1), "USA": (0.0, 0.2)}

print("market     | KS index, 6 dims | KS, original 4 dims | friction (weights 1/1/1) | friction (culture x3)")
rows = []
for c in LAYERS:
    k6, k4 = ks(c), ks(c, range(4))
    f1 = k6 + LAYERS[c][0] + LAYERS[c][1]
    f3 = 3 * k6 + LAYERS[c][0] + LAYERS[c][1]
    rows.append((c, k6, k4, f1, f3))
    print(f"{c:10s} | {k6:16.2f} | {k4:19.2f} | {f1:24.2f} | {f3:21.2f}")
for col, name in ((1, "KS 6-dim"), (3, "friction 1/1/1"), (4, "friction culture x3")):
    print(f"ranking by {name:19s}: " + " < ".join(r[0] for r in sorted(rows, key=lambda r: r[col])))
# market     | KS index, 6 dims | KS, original 4 dims | friction (weights 1/1/1) | friction (culture x3)
# Malaysia   |             2.57 |                1.08 |                     3.67 |                  8.81
# Indonesia  |             1.25 |                1.23 |                     2.85 |                  5.34
# Thailand   |             4.01 |                3.60 |                     4.71 |                 12.74
# Vietnam    |             1.54 |                1.61 |                     2.04 |                  5.13
# Singapore  |             1.35 |                1.21 |                     1.65 |                  4.34
# USA        |             5.64 |                3.55 |                     5.84 |                 17.12
# ranking by KS 6-dim           : Indonesia < Singapore < Vietnam < Malaysia < Thailand < USA
# ranking by friction 1/1/1     : Singapore < Vietnam < Indonesia < Malaysia < Thailand < USA
# ranking by friction culture x3: Singapore < Vietnam < Indonesia < Malaysia < Thailand < USA

# Finding a local partner: the open market versus a referral network (a 'bamboo network' of
# co-ethnic business ties; Weidenbaum & Hughes 1996). Stylised; numbers are assumptions.
#  Open market: 200 candidates; 70 % are reliable. A due-diligence check flags an unreliable firm
#    with probability 0.6 (and wrongly flags a reliable one with probability 0.1). Partner quality
#    (the value created if the partner is reliable) varies across candidates, ~ Normal(10, 3); the
#    firm picks the best-looking unflagged candidate it can find among those it screens.
#  Network: only 20 candidates are reachable, but each comes vouched for by a contact who loses
#    standing in the network if the referral cheats, so the share of reliable candidates is 90 %.
#  An unreliable partner destroys 25 (the firm's specific investments, plus lost time).
import random

def search(pool, reliable_share, screened, rng, flag_bad=0.6, flag_good=0.1, loss=25.0):
    best = None
    for _ in range(min(screened, pool)):
        good = rng.random() < reliable_share
        quality = rng.gauss(10, 3)
        flagged = rng.random() < (flag_good if good else flag_bad)
        if flagged: continue
        if best is None or quality > best[0]: best = (quality, good)
    if best is None: return 0.0, None
    return (best[0] if best[1] else -loss), best[1]

def mean_outcome(pool, share, screened, runs=20000):
    rng = random.Random(7); tot = bad = 0
    for _ in range(runs):
        v, good = search(pool, share, screened, rng)
        tot += v; bad += good is False
    return tot / runs, bad / runs

print("route                                   | candidates screened | mean value | P(unreliable partner)")
for label, pool, share, screened in (("open market", 200, 0.70, 5), ("open market", 200, 0.70, 30),
                                      ("referral network", 20, 0.90, 5), ("referral network", 20, 0.90, 20)):
    v, b = mean_outcome(pool, share, screened)
    print(f"{label:39s} | {screened:19d} | {v:10.2f} | {b:21.1%}")
# The network's closure is also its limit (Uzzi 1997): once the reliable share in the open market
# rises - say, with better courts, credit bureaus and platform ratings - its advantage shrinks.
for share in (0.70, 0.80, 0.90):
    v, b = mean_outcome(200, share, 30)
    print(f"open market with reliable share {share:.0%}, 30 screened: mean value {v:.2f}, P(unreliable) {b:.1%}")
# route                                   | candidates screened | mean value | P(unreliable partner)
# open market                             |                   5 |       6.93 |                 15.7%
# open market                             |                  30 |       9.31 |                 15.9%
# referral network                        |                   5 |      11.44 |                  4.5%
# referral network                        |                  20 |      13.46 |                  4.8%
# open market with reliable share 70%, 30 screened: mean value 9.31, P(unreliable) 15.9%
# open market with reliable share 80%, 30 screened: mean value 11.68, P(unreliable) 10.2%
# open market with reliable share 90%, 30 screened: mean value 13.95, P(unreliable) 4.9%