15Finance, Product Control and Tax

Between mid-2012 and mid-2015 Denmark’s tax authority paid 4 170 claims for refunds of dividend tax, just under 12.1 billion kroner (about £1.4 billion), to claimants whose share trades had been arranged on cum-ex terms. None of the claims was valid under Danish law. The scheme’s principal organiser is among those convicted in Denmark; a New York jury awarded the authority $500 million against some of the parties. But in October 2025 London’s Commercial Court dismissed the authority’s contested civil claims against the defendants there, because it had not shown that it was deceived: its controls for paying refund claims, the judge found, “were so flimsy as to be almost non-existent”. A tax rule shapes which trades exist, and a trade whose profit is a tax refund is a legal question before it is a trading question. So is a P&L number: someone has to sign it, and what they check decides what it means.

15.1 The daily P&L: flash, final and sign-off

Every trading day ends with two numbers. The first comes minutes after the close, from the desk’s own systems and marks; the second the next morning, after every trade is booked, fees are charged and the marks are checked by someone other than the trader (Book 6’s product control).

Built as a sequence of revaluations of the same book, the walk adds up by construction: move the marks from flash to close on the flash trades, add the late trades at the close, charge the fees, then move the marks from close to verified. Each step is the difference of two totals of a firm.pnl book (Book 1), and the steps telescope from the flash to the final (Listing 15.1). A walk that does not add up is itself an exception: some trade, mark or fee is in one number and not the other.

def walk(sod, sod_marks, fills, cut, flash_marks, close_marks, final_marks):
    early = day_book(sod, sod_marks, fills, cut, fees=False)
    all_nofee = day_book(sod, sod_marks, fills, None, fees=False)
    all_fee = day_book(sod, sod_marks, fills, None, fees=True)
    flash = early.total(flash_marks)
    w = {"flash": flash,
         "market moves after the flash": early.total(close_marks) - flash,
         "late trades": all_nofee.total(close_marks) - early.total(close_marks),
         "fees": all_fee.total(close_marks) - all_nofee.total(close_marks),
         "valuation adjustments": all_fee.total(final_marks) - all_fee.total(close_marks),
         "final": all_fee.total(final_marks)}
    w["unexplained"] = w["final"] - w["flash"] - sum(v for k, v in w.items() if k not in ("flash", "final"))
    return w


def exceptions(w, abs_tol, rel_tol, cat_tol):
    """Reasons to stop a sign-off: the flash-to-final difference beyond max(abs_tol, rel_tol |flash|); any
    category beyond its own tolerance; any unexplained remainder."""
    out = []
    diff = w["final"] - w["flash"]
    if abs(diff) > max(abs_tol, rel_tol * abs(w["flash"])):
        out.append(("flash to final", diff))
    for k, tol in cat_tol.items():
        if abs(w[k]) > tol:
            out.append((k, w[k]))
    if w["unexplained"] != 0:
        out.append(("unexplained", w["unexplained"]))
    return out

15.2 Tutorial: a day from flash to final

The final is 20.2% below the flash, $95 967, which is inside the total tolerance of $100 000; a rule on the total alone would sign it. The category rule stops it: valuation adjustments of $149 787 exceed their $100 000 tolerance, so the day is escalated with the reason recorded (fm_finance.day_signoff). The adjustment has one clear cause. The desk closed its largest illiquid position, 73 000 shares, at $123.49 against a consensus of $122.01, 1.2% higher; verification moved the mark to the edge of the 0.5% band, $122.63, and did the same, in smaller amounts, on the other two. A desk whose closing marks on illiquid names sit at the favourable edge of consensus day after day is the pattern product control exists to see.

15.3 Valuation control in a trading firm

Product control’s authority rests on the same independence as the risk function’s (chapter 12): it reports outside the desks, it owns the marks that go into the books and records, and its adjustments cannot be reversed by the trader. In a bank the function is large and regulated; in a proprietary firm or a fund it may be two people, or an administrator (chapter 4), and the temptations are the same. A trader’s marks on illiquid positions are an estimate by the person whose pay depends on it (chapter 10). The controls are Book 6’s: independent verification against consensus or observed trades, tolerance bands, reserves for what cannot be verified, and unexplained P&L investigated rather than absorbed.

15.4 Taxes that shape strategies

Taxes enter a strategy’s return as costs on its trades, on its income and on its gains. The first two are paid whatever the result and scale with activity; they decide which strategies exist in which markets.

For a strategy earning 8% a year, a tax of 10 basis points takes half its return at a turnover of 40 times a year, 20 basis points at 20 times, and 50 basis points, the UK’s rate on share purchases (Box 15.1), at 8 times (Figure 15.2). A market maker or statistical arbitrageur turning its book over forty times a year cannot trade a taxed instrument at 50 basis points; it trades where the tax does not apply, where the rules allow, or not at all.

Withholding tax falls on the other end: the income strategies. A value strategy with a 3.5% dividend yield loses 0.53% a year to an unrecoverable 15% withholding, and a turnover of 0.8 a year costs it only 0.40% at 50 basis points; for a statistical arbitrage book turning over forty times, the same 50 basis points cost 20% a year (Table 15.2). The rates in the table are illustrative inputs; the dated box holds the rules as they stand.

The Danish case is the first kind of trade taken to its end: a trade whose economic content was a tax refund, claimed on shares the claimants had not owned. The UK court found every one of the 4 170 claims invalid under Danish law, and found that the scheme’s architects had no reasonable basis to believe them valid; it still dismissed the contested civil claims for deceit, on the ground that the authority’s processes did not rely on what it was told. Book 1 describes the German cases, which ended in criminal convictions. For a trading firm the lesson is the chapter’s hook: a strategy whose return depends on a tax position is approved by the firm’s tax and legal functions before it is traded, and its profit is audited as a tax position, not a trading one.

With illustrative rates of 37% short-term and 20% long-term, the 60/40 rule taxes a futures gain at a blended 26.8% whatever the holding period, which is why the same trend-following return can be worth more after tax in futures than in stocks held for weeks (firm.signoff.blended).

15.5 Legal entities and transfer pricing

A firm that trades in several countries does so through several legal entities: a trading entity where its traders sit, an entity that holds the exchange memberships, a management company that employs the researchers, a fund or a proprietary vehicle that owns the capital. Each is taxed where it is resident, and each charges the others for what it provides: research, execution, technology, capital.

For a trading firm the hard cases are the ones this book has already met: who owns the strategy (chapter 11), where the decisions that earn the profit are taken, and what a desk in one country pays a research team in another. The OECD’s Transfer Pricing Guidelines (2022 edition) are the reference text on the arm’s length principle; the documentation they require is kept by finance, and it is the reason a firm’s organisation chart and its legal-entity chart are drawn separately and must agree.

15.6 Build: the sign-off engine

15.7 Exercises

15.8 Problem: The Turnover That Tax Takes

15.9 Interview questions