Tracking error in practice: why your results will differ from the model portfolio

Two different comparisons people conflate

There are two distinct tracking questions in copy trading, and they get mixed up constantly.

The first is: does my sleeve match the leader it copies? That is execution tracking error — latency, book depth, order sizing, funding timing — and it is covered in the note on why copy trades never perfectly match the leader.

The second is: does my account match the model portfolio published on the site? That is a different question with different dominant terms, and it is the comparison most users actually make, because the model portfolio number is the one on the screen. This note is about the second.

What the model portfolio is, and what it is not

The model portfolio is a construction. It takes the current basket of scored leaders, assigns each a weight in proportion to their composite score, and computes a weighted return over a window from the leaders' own on-chain history. It is a statement about what the selection and weighting rules produced over that window, using public data anyone can verify against the chain.

It is not a claim about any user's realised results. It has no entry date specific to you, no capital constraint, no fills of its own and no fee drag from your turnover. It also has no path dependency from partial funding: every sleeve in the model is fully weighted for the whole window, which is rarely true of a live account.

Read it as a reference series for the strategy rules, in the same way an index is a reference for a market. Nobody's account equals the index either, and the reasons are the same ones listed below.

Source 1 — entry timing

This is usually the largest term, and it is the one people account for least.

The model portfolio measures a continuous window. You started on a specific day, into a basket that already held open positions at whatever prices those positions were opened. If a sleeve was already 6% into a winning trade when you joined, you did not capture that 6%; you captured whatever happened afterwards, including the possibility of giving back part of a move you were never paid for.

The effect is symmetrical and can help you: joining mid-drawdown means you did not take the drawdown either. But it is large. Over short windows, entry timing can easily dominate every other source on this list combined, which is why comparing your first two weeks against a 30-day model figure produces a meaningless number.

Source 2 — capital size and minimum order sizes

The model assumes weights can be expressed exactly. A real account cannot always express them. Perpetual markets have minimum order sizes and price increments, so a small sleeve rounds: a leader's 0.7% position becomes your 0% or your 1.4%, depending on which side of the minimum you land.

The smaller your capital, the coarser this rounding is, and it does not cancel out cleanly. It systematically drops the smallest positions a leader takes, which changes the strategy you are running — often in the direction of holding fewer, larger positions than the leader does.

Source 3 — Starter Mode versus Full Mode

If you are in Starter Mode you are mirroring one leader. The model portfolio is a weighted basket of up to ten. These are not the same strategy and there is no reason for them to produce similar numbers.

This is worth stating bluntly because it is the most common misread of the performance page. A Starter Mode account should be compared against its own single leader's history, not against the basket. Comparing a one-sleeve account to a ten-sleeve reference will show a much wider dispersion in both directions, and that dispersion is the concentration you are carrying, not a tracking failure.

Source 4 — fill quality, latency and book depth

Every mirrored entry and exit is your own order in the book, placed after the leader's fill was observed. On liquid majors during ordinary conditions the difference is small. During a fast move, in a thin alt perp, or on a large position relative to depth, it is not.

This term is roughly proportional to turnover: a leader who trades ten times a day imposes ten times the exposure to it as one who trades once. Two accounts following different leaders can therefore have very different tracking behaviour for reasons that have nothing to do with either leader's skill.

Source 5 — funding

Funding is paid or received continuously on open perpetual positions, and it accrues to whoever holds the position at each funding interval — which is you, on your own position, at your own size.

The direction matters as much as the magnitude. On the receiving side, funding is a credit that improves your result relative to a price-only comparison. On the paying side it is a persistent drag that grows with holding period. A model figure that treats a window in aggregate will not decompose neatly into your funding experience, particularly if you were not holding for the full window.

Source 6 — fees and the builder fee

Your account pays Hyperliquid's exchange fees on every mirrored fill, plus the builder fee of 0.1% of mirrored notional volume. Both scale with turnover, not with capital or with profit.

This is a knowable, one-directional term: it always subtracts. If you want a like-for-like comparison against a model figure, this is the one component you can estimate reasonably well from your own trade feed, and it is worth doing before concluding that a sleeve is underperforming its leader.

Source 7 — rebalance timing and unfunded sleeves

The model portfolio holds target weights continuously. Your account holds actual weights, which drift as sleeves win and lose, and are only corrected when a rebalance is run and confirmed.

A sleeve that is underfunded relative to its target contributes less than the model assumes, in both directions. If you have not rebalanced in a while, or a readiness blocker prevented a sleeve from being funded at all, your account is running a different weighting than the reference — sometimes very different. This is the most common cause of a gap that looks mysterious and turns out to be arithmetic.

Which sources dominate

Ordering these roughly, from largest expected contribution to smallest, for a typical account over a one-month window: entry timing first, by a distance; then mode mismatch if you are comparing a Starter account to the basket; then unfunded or drifted weights; then fees and funding, which are moderate but persistent; then fill quality, which matters most for high-turnover leaders; then rounding from minimum order sizes, which matters most for small accounts.

The useful implication is that the two largest terms are structural and knowable in advance. If you correct for entry date and mode before comparing, most of the confusing gap disappears.

How to compare correctly

Four practices make the comparison meaningful rather than decorative.

• Use the same window. Compare from your actual start date, not the model's 7D or 30D default.
• Compare like configurations. A Starter account belongs against its single leader's history.
• Compare per sleeve where you can. Aggregate gaps hide the one sleeve that explains them.
• Include cost. Subtract your own fees and funding from the reference before judging the difference.

When a gap is a signal rather than noise

Random divergence changes sign. If your account is sometimes ahead of the reference and sometimes behind, that is normal tracking noise and there is nothing to fix.

A gap that is persistently one-directional over weeks is different, and it usually has a mechanical cause: a sleeve that was never funded, capital sitting in Spot rather than Perps, a readiness blocker that stopped an allocation, weights that drifted far from target, or a much higher turnover leader than the basket average paying more in fees and slippage.

Those are all checkable in the dashboard and in your own trade history. Check them before concluding anything about the strategy.

Conclusion: a model portfolio is a reference, not a promise

A published model portfolio earns its place by being computed transparently from public data with rules stated in advance. It does not earn the right to be treated as a forecast of your account, and no honest presentation of one should imply otherwise.

Expect a gap. Expect entry timing to explain most of it early on. Expect fees and funding to explain a steady, small part of it forever. Investigate only the part that is persistent and one-directional, because that part usually has a cause you can fix.

Model portfolio figures and per-trader history are published on the performance page, so you can compare your own window against exactly the data the scoring uses.