24Entering a New Market

The plan says eight months to the first trade. The clearing agreement alone takes five, and it cannot start until the legal entity exists and has its identifier; the exchange will not certify the software until the clearing firm has set the firm’s limits; and the regulatory registration that membership requires takes four months of its own, starting only when the entity exists. Drawn as a network, the entry takes ten months, and the long pole is not the clearing agreement everyone was watching but the registration nobody had put on the chart.

24.1 Why enter: the edge and the size of the prize

A trading firm enters a new market (an asset class, a region, a venue) because an edge it has elsewhere should work there, or because a client or counterparty it already serves trades there. Flow Traders’ 2025 annual report is a public example of the second kind of reasoning: it described expanding its presence in Asia and “initiating active trading in China, an attractive growth market aligned with our core ETP strengths”, with its exchange-traded product value traded in Asia rising from € 114 billion in 2024 to € 152 billion in 2025. Whatever the reason, the entry costs money for months before the first trade, and the prize is uncertain until the firm has traded for a while.

24.2 The checklist: access, clearing, data, legal, staffing

The workstreams are the chapters of this series. The entity needs a legal entity identifier (Book 2) before most counterparties will open an account; authorisation or registration follows chapter 17’s map and its statutory periods; clearing through a futures commission merchant or a clearing member (Book 1) needs know-your-customer checks, legal agreements (chapter 18) and limits; membership and certification (Book 13) follow; connectivity (Book 14), market data licences (chapter 22) and people (chapter 10) run alongside. The chapter’s example is a futures market in a new region (Table 24.1); durations and costs are illustrative inputs.

24.3 The critical path

Kelley and Walker described the method in 1959 for construction and maintenance projects; it applies unchanged to an entry plan (Listing 24.1).

def schedule(tasks):
    by = {t.name: t for t in tasks}
    es, ef = {}, {}

    def finish(n):
        if n not in ef:
            t = by[n]
            es[n] = max((finish(p) for p in t.after), default=0.0)
            ef[n] = es[n] + t.months
        return ef[n]

    end = max(finish(n) for n in by)
    lf = {n: end for n in by}
    for n in sorted(by, key=lambda x: -ef[x]):          # latest finish: min over successors' latest start
        for s in by.values():
            if n in s.after:
                lf[n] = min(lf[n], lf[s.name] - s.months)
    return {n: (es[n], ef[n], lf[n] - ef[n]) for n in by}


def critical_path(tasks):
    """Tasks with zero slack, in order of their earliest start."""
    sch = schedule(tasks)
    return [n for n, (s, _, sl) in sorted(sch.items(), key=lambda x: (x[1][0], x[1][1])) if abs(sl) < 1e-9]


def cost_by_month(tasks, horizon):
    sch = schedule(tasks)
    out = np.zeros(horizon)
    for t in tasks:
        s, f, _ = sch[t.name]
        for m in range(horizon):
            overlap = max(0.0, min(f, m + 1) - max(s, m))
            out[m] += overlap * t.monthly_cost
    return out

The critical path runs through the legal entity, the regulatory registration, the exchange membership and the certification: ten months to the first trade (Figure 24.1). The clearing chain, which the plan’s eight months were built around, finishes after eight months and has a month of slack. Connectivity has four months of slack and the data licences six: they can start late without cost. Before the first trade the entry spends $949 000, at most $170 000 in a month, when software, hires, registration and connectivity overlap.

24.4 Staging the investment

After the first trade the revenue ramps towards a steady state $R$ a month, as $R(1-e^{-t/\tau})$ with $\tau=4$ months, against running costs of $250 000 a month; the horizon is three years and the discount rate 10%. The steady state is the uncertain part: in the chapter’s model it is lognormal with a median of $400 000 a month. A weak outcome ($150 000) loses $3.55 million of present value; the base case ($450 000) gains $2.00 million, an internal rate of return of 104% a year; a strong one ($800 000) gains $8.46 million.

In practice the firm sees a noisy estimate. The chapter’s firm observes $R$ with a 25% error after six months of live trading and stops if the estimate is below the $250 000 running cost (Listing 24.2).

def staged_npv(tasks, first_trade, Rs, tau, run_cost, horizon, rate, kill_after, threshold, noise, seed=0):
    """For each true steady-state revenue in Rs: the NPV without staging, and with a kill point after `kill_after`
    months of trading, where the firm observes R with multiplicative noise and stops if the estimate is below
    `threshold`. Returns (npv without, npv with, killed flags)."""
    rng = np.random.default_rng(seed)
    stop_month = int(math.ceil(first_trade)) + kill_after
    plain, staged, killed = [], [], []
    for R in Rs:
        plain.append(npv(entry_cash(tasks, first_trade, R, tau, run_cost, horizon), rate))
        est = R * math.exp(noise * rng.standard_normal())
        k = est < threshold
        killed.append(k)
        staged.append(npv(entry_cash(tasks, first_trade, R, tau, run_cost, horizon, stop_month if k else None), rate))
    return np.array(plain), np.array(staged), np.array(killed)

24.5 Tutorial: eight months to first trade

Without a kill point the entry’s expected NPV is $2.52 million, with a 38.6% chance of losing money and a 5th percentile of $-\$3.57$ million. With the kill point the firm stops 22.7% of entries; the expected NPV rises by $200 000 to $2.72 million, and the 5th percentile to $-\$1.92$ million. The probability of losing money barely moves (39.6%): stopping turns large losses into smaller ones rather than into gains, and the set-up costs are sunk. The noise costs something: 3.6% of all entries are weak and continue, and 5.5% are sound and stopped.

The threshold matters as much as the kill point (Figure 24.3). Set at the running cost, as Proposition 24.4 suggests, it adds $200 000; at $300 000 it adds $150 000; at $400 000 it destroys $138 000, because a rule that demands the base case kills entries that would have paid. A kill criterion is written before entry, at a level the firm could defend if the market then turned out well.

24.6 A worked entry plan

24.7 Build: the entry plan

24.8 Exercises

24.9 Problem: Eight Months to First Trade

24.10 Interview questions