comparing-etf-tracking-difference-and-error — evaluating investments

The practical point: if your ETF lags its index by 0.10% per year, you are accepting a cost that is roughly $2,000 per $100,000 over 10 years at 7%—and that drag compounds whether you "feel" it or not.

Why ETF Tracking Difference and Error Matter

If you hold an ETF for 20 years, you experience 2 different risks: a persistent return gap measured in basis points (bps), and a variability of that gap measured as an annualized percentage.

If you mix them up by 0.50% vs 0.05%, you can pick a fund that is "stable" yet still underperforms by 50 bps every year—which the math treats as a guaranteed headwind rather than "noise" (see the $100,000 / 10-year example under "Common Implementation Mistakes").

Definitions (With Numbers You Can Audit)

Tracking difference (TD): "How much did you actually lag?"

Tracking difference is the annualized return difference between the ETF and its benchmark over a stated window like 5 years.

In the dataset's S&P 500 example, SPY shows -0.09% TD with an expense ratio of 0.0945%, while IVV and VOO show -0.03% TD with 0.03% expense ratios.

If TD is -0.09%, you are giving up 9 bps per year versus the index, which is $90 per $100,000 per year before compounding.

Tracking error (TE): "How noisy was the gap?"

Tracking error is the volatility (standard deviation) of the ETF-minus-index return differential, typically annualized from monthly differentials.

In the same S&P 500 trio, the dataset reports 0.12% TE for SPY, 0.08% for IVV, and 0.07% for VOO.

A TE difference of 0.12% vs 0.07% is a 5 bps volatility difference, which matters far more for tactical 3–12 month trades than for 20-year compounding.

Quantified Rules and Thresholds (Use These, Not Vibes)

Tracking difference quality (relative to fees)

Treat TD relative to the expense ratio using these thresholds:

Excellent: TD is within 0.05% of the expense ratio.
Acceptable: TD is within 0.15% of the expense ratio.
Poor: TD is >0.25% beyond the expense ratio.

A gap of >25 bps beyond fees is a clue that something structural (like sampling, trading costs, cash drag, or roll costs) is eating returns by more than 0.25% per year.

Tracking error benchmarks (annualized)

Use these TE "sanity bands":

US large-cap equity: <0.10%
US small-cap equity: <0.25%
Developed international: <0.50%
Emerging markets: <1.00%
US aggregate bond: <0.30%
High-yield bond: <0.75%

Evaluation windows (so your statistics mean something)

Minimum: 36 months of returns; preferred: 60 months including at least one >10% drawdown.

A 60-month window that includes a >10% correction gives you at least 1 stress regime where costs and market microstructure can widen gaps.

What Actually Causes TD and TE (With Measured Magnitudes)

Expenses and cash drag are not "small" at 28 bps

From 1993–2001, SPY underperformed the S&P 500 by 28 bps annually, with 18 bps tied to expenses and 10 bps tied to dividend cash drag (Elton, Gruber, Comer & Li, 2002).

If your index return is 8%, a recurring 0.28% shortfall is 3.5% of your gross return each year.

Securities lending can offset 1–15 bps (but only if you count it)

Some ETFs offset 0.02%–0.15% of expenses via securities lending, and the dataset notes a case where VTI's ~0.01% offset meaningfully changes a fee comparison by 1 bp vs 2 bps.

Vanguard's reported $68 million of securities lending revenue in 2022 averaged about a 0.005% offset across equity funds, which is 0.5 bps that you either capture or ignore.

Liquidity and arbitrage efficiency show 0.15% vs 0.52% TE

In one empirical result, high-volume ETFs had tracking errors averaging 0.15% annually, while low-volume ETFs averaged 0.52% annually, a 3.5× difference (Petajisto, 2017).

The mechanism is mechanical: more frequent creation/redemption reduces NAV-price divergence by tens of bps when markets are stressed (Petajisto, 2017).

International and emerging markets can push TE above 1.00%

International equity ETFs show mean TE of 1.42% annually versus 0.38% for domestic equity ETFs, with emerging markets reaching 2.18% (Rompotis, 2011).

The driver is the time-zone gap plus fair value pricing, which can add 0.5%–1.2% annualized "artificial" TE if you benchmark against local closes instead of NAV-aligned measures.

Bonds can run large negative TD (especially corporate credit)

Bond ETFs averaged -0.72% TD annually versus -0.21% for equity ETFs, and corporate bond ETFs hit -1.14% (Buetow & Henderson, 2012).

A -1.14% structural gap is 114 bps per year, which dominates a 30–75 bps TE guideline for credit if your goal is long-horizon replication.

Worked Example: You Compare S&P 500 ETFs Like an Adult (Numbers First)

You start with $250,000 earmarked for 20 years, and you assume the S&P 500 index earns 8% annually.

You evaluate SPY vs IVV vs VOO using a 5-year TD window, then a TE window, then trading costs.

Step 1: You compute tracking difference versus the index

You record the dataset's TD and fee pairings: SPY -0.09% TD (0.0945% fee), IVV -0.03% TD (0.03% fee), VOO -0.03% TD (0.03% fee).

You immediately see a 6 bps/year TD penalty from SPY versus IVV/VOO (9 bps vs 3 bps).

Step 2: You check tracking error for consistency

You note TE is 0.12% (SPY), 0.08% (IVV), 0.07% (VOO).

You treat the 0.05% TE spread (12 bps vs 7 bps) as a short-horizon risk metric, not a long-horizon return metric.

Step 3: You price the one-time trading friction

You compare liquidity: SPY 85 million shares/day with a 0.003% spread, IVV 5 million/day with 0.008%, VOO 4 million/day with 0.010%.

On $250,000, a 0.010% spread is roughly $25 one-time, which is smaller than 1 year of a 6 bps TD gap ($150/year on $250,000).

Step 4: You translate bps into 20-year dollars

You compare terminal wealth outcomes provided in the dataset: a baseline selection (VOO or IVV) reaches $1,145,936 at 7.97% net, while selecting SPY yields $1,125,112 at 7.91% net.

You find the cost of the suboptimal selection is $20,824 over 20 years, and the point is that 6 bps/year can still compound into 5 figures.

Step 5: You choose based on holding period

For a 20-year buy-and-hold, you minimize TD first, because the spread cost is a 0.003%–0.010% one-time event while TD is a 0.03%–0.09% recurring annual drag.

If you also model a good scenario where lending offsets 0.01%, the dataset's terminal value rises to $1,150,024, a $4,088 gain versus baseline.

Historical Examples Where TD/TE Became the Whole Trade (Exact Dates, Exact Gaps)

April 2020–June 2020: USO vs spot oil (62 percentage points missing)

From April 2020 to June 2020, spot WTI recovered 85% from April lows while USO recovered 23%, a 62 percentage point shortfall (Bloomberg Terminal; USO prospectus supplement filed May 2020).

In May 2020 alone, USO lost an additional 8.7% to contract rolling, which is a single-month cost bigger than a 0.30% annual TE guideline by roughly 29×.

August 24, 2015: EEM discount hit -4.8% intraday (then normalized in 3 days)

On August 24, 2015, EEM traded at a -4.8% discount to NAV, versus typical -0.02% to +0.02%, which is a widening of roughly 4.78–4.82 percentage points (BlackRock iShares NAV data; SEC Market Structure Report Dec 2015).

If you sold into that discount, you locked in a loss 4.8% larger than portfolio value, and the spread normalized within 3 trading days.

March 9–March 23, 2020: BND discount hit -6.2% and TE spiked to 2.4%

From March 9, 2020 to March 23, 2020, BND hit a -6.2% discount to NAV on March 12, versus a 5-year average of -0.01%, a gap of about 6.19 percentage points (Vanguard NAV reports; Fed announcement March 23, 2020).

During that window, TE spiked to 2.4% annualized versus a 0.18% historical average, then recovered to 0.22% by April 15, which is a ~13× spike that later mean-reverted to within 4 bps of normal.

Common Implementation Mistakes (And Their Quantified Costs)

Mistake 1: You treat 0.05% TE as "good" and ignore -0.50% TD

If you buy an ETF with 0.05% TE but -0.50% TD, you underperform by 50 bps every year even though the deviation is "stable."

On $100,000 over 10 years at a 7% index return, the dataset quantifies the terminal wealth loss as $9,847, which is the cost of confusing volatility (5 bps) with drift (50 bps).

Mistake 2: You ignore securities lending and mis-rank by 1–2 bps

If you compare 0.03% vs 0.04% expense ratios but ignore a ~0.01% lending offset, you think the gap is 1 bp when it is actually 2 bps in the dataset's example.

Over 20 years, a persistent 1 bp mismeasurement can become hundreds to low thousands of dollars on $250,000, because 0.01% is still $25/year before compounding.

Mistake 3: You compare international TE to US TE without adjusting the benchmark clock

If you compare an international ETF TE of 0.65% to SPY's 0.12% and call the 0.53% gap "manager inefficiency," you can be overstating inefficiency by about 0.35% due to fair value pricing artifacts.

That 35 bps misread is larger than an "excellent" TD band of 0.05%, so the ranking can flip purely due to measurement error.

Implementation Checklist (Tiered by ROI)

Highest ROI (do these in 30–60 minutes)

Compute TD over 60 months (preferred) and at least 36 months (minimum), and label results in bps/year.
Flag any fund where TD is >0.25% beyond its expense ratio as poor, and treat >25 bps as a structural warning.
Convert TD into dollars using the dataset's rule: each 0.10% TD costs about $2,000 per $100,000 over 10 years at 7%, then scale to $250,000.

Medium ROI (do these if you trade or buy international/bonds)

Check TE against the right asset-class band (e.g., <0.10% for US large-cap vs <1.00% for EM) so you don't misclassify a 2.18% EM TE as "broken" when it's "structurally high" (Rompotis, 2011).
For bonds, treat -0.72% TD averages and -1.14% corporate TD as baseline expectations that can dominate a 0.30% TE guideline (Buetow & Henderson, 2012).
For international ETFs, compare against NAV-aligned or peer benchmarks to avoid a 0.5%–1.2% annualized artifact in TE.

Lower ROI (useful, but only after TD/TE are clean)

Compare bid-ask spreads like 0.003% vs 0.010% and decide if the one-time cost is smaller than 1 year of TD (as it is in the $250,000 example).
Use the break-even rule: if you save 0.10% TD but pay $10 more in commission, break-even on $10,000 occurs at 1.0 years.

The Durable Lesson

Tracking difference is drift (e.g., -0.09% vs -0.03%) and tracking error is noise (e.g., 0.12% vs 0.07%), and you price drift over 20 years in $20,824 chunks while you price noise over days to months in percent-discount events like -4.8% (2015-08-24) or -6.2% (2020-03-12).

Cited sources:

Petajisto, A. (2017). Inefficiencies in the Pricing of Exchange-Traded Funds. Financial Analysts Journal, 73(1), 24–54. https://doi.org/10.2469/faj.v73.n1.7
Elton, E.J., Gruber, M.J., Comer, G., & Li, K. (2002). Spiders: Where Are the Bugs? Journal of Business, 75(3), 453–472. https://doi.org/10.1086/341849
Rompotis, G.G. (2011). Predictable Patterns in ETFs' Return and Tracking Error. Studies in Economics and Finance, 28(1), 14–35. https://doi.org/10.1108/10867371111110534
Buetow, G.W. & Henderson, B.J. (2012). An Empirical Analysis of Exchange-Traded Funds. Journal of Portfolio Management, 38(4), 112–127. https://doi.org/10.3905/jpm.2012.38.4.112
Historical datasets: Bloomberg Terminal (USO, Apr–Jun 2020), USO prospectus supplement (May 2020), BlackRock iShares NAV data (EEM, 2015-08-24), SEC Market Structure Report (Dec 2015), Vanguard daily NAV reports (BND, Mar 2020), Federal Reserve announcement (2020-03-23).

Comparing ETF Tracking Difference and Error