When I utter the words “trading costs” some of you will just yawn, some will have already fallen asleep. [alarm clock][or air horns for the back row] – don’t, just yet; because it matters for your strategy more than you think. For any strategy, from the ones run by Jane Street PMs to those of Günter — my relentless next door neighbor trying to qualify for his 10th prop shop challenge just this year.
The Jane Streets, Citadels, Millenniums of this world have advantages that most of us are looking at with awe and horror, with their:
thousands of math camp winning quants,
nano-second enabled infrastructure,
all but unlimited capital,
pooled signal and risk layers upon layers,
probably flying robots fixing their coffee machine before it’s even broken...
Little attention is paid to the dull “top-tier” trading costs simply because they don’t evoke any sci-fi movie scene.
They were able to build this arsenal, however, partly because they were “born into” this last tier. Once conferred, these cost levels allowed them to compete mano-a-mano, elite-to-elite. You and I on our own, though, need to start in the decrepit commission-slums of this unbalanced world.
So I designed – ok, prompted into existence – a little toy strategy1 to demonstrate this hereditary advantage. The inspiration came from this paper, but as would be expected with my background, I picked the mean reversion model and shortened the time horizons. How could I not?? A month?! I didn’t even know that timeframe existed in trading... ha...
You can download and evaluate it here, and I think it may be useful beyond this exercise. Full method, verified fee schedules, and code are in the gist — this is a demo on free data with stated assumptions, disclaimers, the lot.
I will cut to the chase. But just before that, two important points.
The 85% limit-order assumption23 came from the same Frazzini paper that provided the the other half of the inspiration. It’s the share of orders sent as limits by the institutions in that study, so I used the same. The rest are assumed aggressive.
The κ=0 assumption means limit orders execute at the mid, market orders at the opposite side. That limit-execution point is more conservative than in the paper; Top-tier institutions, as we all know, have a super-machine that would dodge most of the adverse selection — but I couldn’t assume it away entirely either, for reasons I get into in my earlier article on adverse selection.
The important caveat: I gave your average Trader Joe access to the same execution machine as the institution, unrealistic as that may be (very!). Simple reason — I didn’t want to conflate the cost difference that comes from “position” with the one that comes from something you had to build. So the difference between these strategies is in the fees only.
The institution would turn each million into $3.57 million. Joe, running the very same strategy, would be down more than a quarter. And if Joe has any sense — and doesn’t fancy telling his dad he lost that slice of the family house — he chucks this trading idea in the bin right with the hundreds of others.
I reiterate, the difference was only in commissions and rebate tiers. Not the sexiest words in the English dictionary, I know, but ones that ought to make it into your vocabulary fast.
Before I claim the laurels here, of course I’m aware I’m not the first to point out that trading costs matter, nor the first person to measure them. You could say in all fairness that others have done it more thoroughly, even. Both Novy-Marx & Velikov and Frazzini and co. have, using deep, extensive data and a whole basket of strategies.
In my defense, this is a Substack article — believe me, I’d have done more for arXiv, too... Plus I’d have added at least two more contributors! Yes, the reference tables just ring better that way. “ER = EPR” on wormholes sounds catchier than ER alone, wouldn’t you agree? Poor Podolsky, doing all that work for a middle initial. (If you don’t actually know what I’m rambling on about, believe me, that is normal, and a great relief for me. It means at least one normal person is reading my articles)
But let’s look at what those two papers measured, and why they reached such different conclusions; and arrived at extremely different numbers, too!
Novy-Marx & Velikov measured the cost faced by a retail-scale trader. Frazzini-Israel-Moskowitz — now you get my three-name envy?? — measured what a top-tier institution realistically pays. No need to cagefight this out, because they actually agree on why the numbers differ so much: NMV’s are the costs of the small liquidity demander , Frazzini’s the costs of the wise trading giant4. An order of magnitude apart — and yet both authors shook on it before any foul was called.
So, did I just replicate that study slum-style? Not really. What’s different in the Frazzini paper is that the whole cost stack is collapsed into one number — commission, spread, market impact, adverse selection, financing, opportunity cost, all of it. It bakes in the institution’s ability to rest limit orders 85% of the time, mostly sidestepping adverse selection. And adverse selection is not a penny: in modern markets it between 50% and 80% of the spread. A capitalized execution advantage — which, to Joe, reads as hieroglyphics.
So of course the real top tier pulls even further ahead than my two equity lines suggest. But the part I wanted to point to (even without the comfort of the three initials) is that the extra distance I don’t include is intrinsically different: it is earned. It is the moat these institutions built out for themselves, stone by stone, brick by brick. The distance in my demo, however, is different. It is conferred — it’s the fee schedule, nothing else. The giants are born into that world, and the extra rebates and reduced costs are served to them on a silver platter. And I think it is a failure of the industry to so rarely discuss this side of the advantage.
One more point: the gap I did measure holds whatever the spread turns out to be.
For those about to call out “but on Robinhood I don’t pay any commission!” — yes, and you also can’t run this strategy. I’d love to get into the minutia of the payment-for-order-flow business, with all its mechanics and effects, I really would. But I don’t want to muddle things up, and will leave that conversation for another day. Enough said that once you consider all the other costs you would end up way below even the yellow line above if you tried.
Instead of a witty close or a call-back, I will leave you in parting with the fee summary & comparison of a few popular symbols across some markets, broken down into top-tier player costs vs “the rest”56:
The strategy is a one-day cross-sectional reversal on the S&P 500: each day, long the prior day's decile of biggest losers, short the decile of biggest winners, equal-weighted, one dollar long against one dollar short per dollar of capital, signal at the close and execution at the next open. Construction follows the short-term reversal specification in Novy-Marx & Velikov (2016). Full method, code and data notes are in the gist.
The 85% passive fill rate is based on the institutional execution described in Frazzini, Israel & Moskowitz (2018). κ = 0 means a resting limit order keeps none of the half-spread after adverse selection (it fills, effectively, at the mid), and a marketable order pays half the spread; this is the conservative end for the passive trader. Market impact is not modelled, for either tier.
Spreads are estimated from daily high/low prices using the Corwin & Schultz (2012) estimator, floored at a one-tick and smoothed on a 20-day median. On the names actually traded (the day's biggest movers) this produces ~28 bps; the piece also reports a constant 5 bps case standing in for tick-data quoted spreads.
The 20–57 bps figure is Novy-Marx & Velikov's estimate of trade-execution costs for mid-turnover anomalies. Frazzini, Israel & Moskowitz put a large institution's realised costs an order of magnitude lower — a median of roughly 6 bps per rebalance, and an effective spread of ~1.5 bps against ~21 bps quoted — validated out-of-sample against the realised costs of live index funds (Vanguard, iShares).
Fee figures read September 2026. Equities: IBKR Pro/Lite retail rates versus a direct exchange member paying the schedule directly; the maker rebate reaches $0.0037 combined at MEMX, $0.00325 at Nasdaq. Futures: exchange fee only (member vs non-member), excluding the broker's own commission, NFA and give-up. Options: Cboe fee by capacity (Customer vs Market-Maker sliding scale), excluding broker commission and SPX surcharges; the top market-maker tiers require a $2.4 million annual prepayment. Crypto: Binance is the venue and the counterparty, so the fee is the all-in cost. Full schedules and sources in the [gist]
In IBKR's own words: it "may receive enhanced rebate payments as a result of exceeding volume thresholds on particular markets, but typically will not directly pass these enhancements to customers," and separately, on futures, "may receive aggregate volume discounts that are not passed on to clients." The volume-tier advantage is captured by the aggregator; the individual gets the base rate.
Bibliography
Corwin, S. A., & Schultz, P. (2012). “A Simple Way to Estimate Bid-Ask Spreads from Daily High and Low Prices.” The Journal of Finance, 67(2), 719–760. https://doi.org/10.1111/j.1540-6261.2012.01729.x
Chen, A. Y., & Velikov, M. (2023). “Zeroing In on the Expected Returns of Anomalies.” Journal of Financial and Quantitative Analysis, 58(3), 968–1004. https://doi.org/10.1017/S0022109022000874
Frazzini, A., Israel, R., & Moskowitz, T. J. (2018). “Trading Costs.” Working paper. SSRN: https://ssrn.com/abstract=3229719 · https://doi.org/10.2139/ssrn.3229719
Frazzini, A., Israel, R., & Moskowitz, T. J. (2012). “Trading Costs of Asset Pricing Anomalies.” Fama-Miller Working Paper. SSRN: https://ssrn.com/abstract=2294498 · https://doi.org/10.2139/ssrn.2294498
Kyle, A. S. (1985). “Continuous Auctions and Insider Trading.” Econometrica, 53(6), 1315–1335. https://doi.org/10.2307/1913210
Novy-Marx, R., & Velikov, M. (2016). “A Taxonomy of Anomalies and Their Trading Costs.” The Review of Financial Studies, 29(1), 104–147. https://doi.org/10.1093/rfs/hhv063 · NBER working paper: https://www.nber.org/papers/w20721
Perold, A. F. (1988). “The Implementation Shortfall: Paper versus Reality.” The Journal of Portfolio Management, 14(3), 4–9. https://doi.org/10.3905/jpm.1988.409150
I know that's not the point
But I can't help thinking that Average Joe could have bought the Spy and forgotten about it to make more money than the big fund