Task 2 — Client Profitability & Spread Recommendation
Nomura Quant Challenge 5 | Market Making
1. Client Classification with Quantitative Justification
Each client is classified as profitable or costly based on the sign of their expected aggregate PnL per trade (Eq. 6). The
aggregate PnL averages the LP's realised PnL across all six closing horizons using uniform weights wi = 1/6:
Aggregate PnL = side × V × (1/6) × Σ(Mτ − TP) for τ ∈ {5,10,15,20,25,30}s
A positive aggregate PnL means the mid price moves in the LP's favour on average after execution, indicating the client does
not trade on directional information that systematically hurts the LP. A negative value indicates the client is an informed trader
whose flow generates sustained adverse price moves.
Client τ=5 τ=10 τ=15 τ=20 τ=25 τ=30 Agg PnL δ* Class
Profitab
A +1.268 +1.540 +1.737 +2.061 +2.293 +2.485 +1.8970 0.0000
le
Profitab
B +1.211 +1.342 +1.542 +1.726 +1.909 +2.059 +1.6320 0.0000
le
Profitab
C +1.077 +1.165 +1.190 +1.162 +1.229 +1.216 +1.1730 0.0000
le
Profitab
D +0.868 +0.682 +0.493 +0.177 -0.076 -0.347 +0.2990 0.0096
le
E +0.653 +0.303 -0.108 -0.572 -1.069 -1.614 -0.4010 0.0190 Costly
F +0.445 -0.176 -0.917 -1.751 -2.614 -3.629 -1.4400 0.0329 Costly
Table 1. Expected PnL per trade at each horizon and aggregate PnL (all values from LP perspective). Positive values mean the LP earns
money on average.
Clients A, B, C, D: Profitable. All four have positive aggregate PnL. Client A is the most profitable (agg = +1.897), with
per-horizon PnL that is positive at every horizon and actually grows with τ — the price continues to drift in the LP's favour over
time. Client B (+1.632) and C (+1.173) show a similar pattern. Client D (+0.299) is marginally profitable: per-horizon PnL is
positive at τ=5 (+0.87) but turns negative at τ=25 and τ=30, meaning D's trades become adverse over longer horizons. The
positive aggregate only survives because the early horizons dominate.
Clients E and F: Costly. Client E has aggregate PnL = −0.401 and Client F = −1.440. For both, per-horizon PnL is positive at
τ=5 but turns negative by τ=15 (E) or τ=10 (F), confirming that the adverse price move is not immediate but accumulates
rapidly. Client F is the most toxic: by τ=30 the LP loses −3.629 per unit on average, more than eight times the aggregate loss
per trade.
2. Minimum Half-Spread δ* and Its Relation to the Adversity Profile
Derivation of δ*.
If the LP quotes at M0 ± δ instead of the historical trade price TP, the execution price becomes TPnew = M0 − side × δ.
Substituting into the aggregate PnL formula:
E[side × V × (avg(Mτ) − TPnew)]
= E[side × V × (avg(Mτ) − M0)] + E[V] × δ
= BaseAggPnL + E[V] × δ
Setting this ≥ 0 and solving: δ* = max(0, −BaseAggPnL / E[V]).
For clients whose BaseAggPnL is already positive (A, B, C), δ* = 0 — the LP is profitable even at the current quoted spread. For
client D, the base PnL evaluated at M0 is slightly negative, requiring a minimal δ* = 0.0096. For E and F, larger half-spreads are
needed: δ* = 0.0190 and δ* = 0.0329 respectively.
Connection to the Adversity Profile (Task 1).
The adversity profile measures how often and how quickly the mid price moves against the LP after execution. δ* is the direct
monetary consequence of that adverse movement: it is precisely the spread the LP must charge to break even against each
client's pattern of adverse selection.
Client τ=5 (%) τ=10 (%) τ=15 (%) τ=20 (%) τ=25 (%) τ=30 (%) δ* (units)
A 39.72 41.40 42.19 41.95 41.90 41.71 0.0000
B 40.75 42.36 43.16 43.30 43.42 43.30 0.0000
C 41.37 43.35 44.25 44.84 45.39 45.92 0.0000
D 43.32 46.52 48.17 49.71 50.62 51.75 0.0096
E 44.96 48.49 50.92 52.13 53.85 54.84 0.0190
F 47.01 51.67 54.26 56.94 59.49 61.88 0.0329
Table 2. Adversity profile (% of adverse trades per horizon) alongside δ* for each client.
Three relationships are visible:
Monotone ordering: δ* is ordered identically to adversity at τ=30 — F (61.88%, δ*=0.033) > E (54.84%, δ*=0.019) > D
(51.75%, δ*=0.0096) > C, B, A (all ≤45.92%, δ*=0). Clients with higher adversity at long horizons require larger break-even
spreads.
Slope matters, not just level: The adversity increment from τ=5 to τ=30 — the speed at which trades become adverse —
drives the magnitude of δ*. Client F's adversity rises by 14.8 pp over 25 seconds, which translates into a rapidly worsening
expected PnL across horizons (from +0.45 at τ=5 to −3.63 at τ=30). This sustained adverse drift is exactly what forces δ* to be
large: a wider quoted spread is needed to earn enough up-front to offset the losses that accumulate over the holding window.
Flat profiles need no adjustment: Clients A and B have nearly flat adversity profiles (rises of only 2–3 pp from τ=5 to τ=30)
and have positive expected PnL at every horizon. Their flow carries no sustained directional information, so the current spread
is already sufficient and δ* = 0.
In summary, δ* is a translation of the adversity profile into spread units: it is the minimum premium the LP must embed in the
quoted price to convert a client's adverse selection cost into a break-even position. A steeply rising adversity profile (Clients E,
F) implies a larger δ* because the LP must compensate for adverse price drift that persists over the entire 30-second holding
window, not just at the moment of execution.