Are Thematic ETFs Undervalued or Overvalued? Inside the Valuation Radar

A note on authorship: The research, analysis, and opinions in this article are the author's own. Claude (Anthropic's AI) assisted with drafting and editing the prose.

Are any thematic ETFs undervalued right now? It's a harder question than it sounds, because a thematic ETF isn't one thing — it's a basket of dozens or hundreds of individual stocks, each with its own price and its own intrinsic value. The DCF Valuation tool already answers this for a single stock: on conservative assumptions, what is this business worth, and is it trading above or below that? The Thematic ETF Valuation Radar asks the same question of every stock inside a fund at once, rolls the answer into one number for the whole fund, and does it once a month across a growing list of thematic funds.

Why thematic funds, and not VAS or IVV

The obvious objection: why not run this on a plain S&P 500 or ASX 200 index fund? Because a broad index fund reconstitutes — its rules automatically drop a constituent once it no longer qualifies (too expensive relative to market cap thresholds, too small, delisted) and add the next one in. That self-cleansing means a point-in-time look-through valuation of a broad index has almost no predictive teeth; it can read "overvalued" for a decade while the underlying companies — and the index itself — keep compounding, because the index simply carries whatever the market currently rewards.

A thematic fund can't do that. If you buy a cybersecurity ETF, it stays a cybersecurity ETF — it can't quietly rotate into cheaper, unrelated businesses just because the theme has run hot. That constraint is what makes a look-through measurement mean something: the fund is genuinely locked to whatever its holdings are worth, for better or worse.

How the valuation is actually built

The pipeline is deliberately unglamorous — no new valuation model, just the existing one pointed at a lot more companies at once:

• Real holdings, not a shortcut. Each fund's holdings come from the issuer's own published full-holdings file — BetaShares' and iShares' portfolio CSVs — rather than a data aggregator's ETF holdings feed, which is frequently wrong or stale for an ASX-listed cross-listing of a US fund.
• One valuation per company, however many funds hold it. Across the whole radar, the underlying holdings overlap heavily — Nvidia turns up in a semiconductor fund, a robotics fund, and a software fund all at once. Rather than valuing it three times, every fund's holdings are pooled into one union of roughly 830 unique companies, and each is valued exactly once.
• The same engine as everywhere else on the site. Every holding runs through the identical dispatch the DCF tool and the six undervalued screens use — a five-year discounted cash flow by default, a dividend or excess-return model for banks, insurers and regulated utilities, and an EV/Sales comparison for the small tail of companies the cash-flow models reject outright.
• Weight, then sum. Each holding's premium or discount to its own intrinsic value is weighted by how much of the fund it actually represents, then summed into one figure: the fund-weighted average premium or discount across everything we could value.
• A coverage gate. A fund is only ranked once the holdings we could value cover at least 70% of its total weight — below that, the number would be answering for a fund the model barely saw.
• A sanity band. Any single holding whose output lands outside roughly −200% to +100% is treated as unvaluable and dropped from the blend rather than distorting it — the next section is exactly why that guardrail exists.

Two checks built into every valuation

A look-through number is only as good as the growth assumption behind each individual holding, so every company is put through two checks before it's included in a fund's result.

Currency matching. Some companies report their underlying financial results in one currency but trade in another — an Indian company's US-listed shares, for example, report profits in rupees while the share price is in dollars. Before valuing any holding, its financial results are converted into the same currency as its share price. Skip this step and a currency mismatch alone can make an ordinary business look wildly more valuable than it is.

A multi-year growth trend, not a single year. A common way to estimate a company's future growth is to look at its most recent one-year figure. But a single year can be misleading — a company recovering from a one-off hit, such as a lawsuit or a bad quarter, can post a sharp rebound that isn't a repeatable growth rate. Projected forward for five years, that one unusual year can make an otherwise fairly-priced company look dramatically undervalued. So instead, each holding's growth input is the slower of its multi-year revenue trend and its multi-year free-cash-flow trend — around four years of history — so one unusual year can't drive the whole projection. UnitedHealth is a real example: its earnings had rebounded sharply off a litigation-depressed prior year, which on a one-year view looked like strong growth. On the multi-year trend, its free cash flow was actually shrinking — and it now reads close to fair value rather than dramatically undervalued. A genuinely fast-growing company's premium still shows up either way; what changes is that a company merely recovering from a rough year isn't mistaken for one.

What it says today

As at the 2 September 2026 snapshot, every fund in the radar reads a premium to its combined look-through intrinsic value — none currently reads as a discount. They range from a cloud-computing fund trading closest to fair value, to a semiconductor fund trading furthest above it:

"vs intrinsic value" is the fund-weighted average premium of the fund's holdings to their own combined intrinsic value — a larger negative number means the holdings collectively trade further above what the models say they're worth. Full ranking, prices and implied fair values at the live page, which updates as prices move.

The pattern reads sensibly rather than randomly. The two healthcare funds sit closest to fair value — a sector the market currently prices with less optimism. The semiconductor fund sits furthest out, which is exactly what you'd expect from a five-year, growth-capped model pointed at the middle of an AI capital-spending run. That's not a coincidence — it's the next section's point made concrete.

What the premium actually means, and what it doesn't

Nothing here is a buy, sell, or hold call — that's a legal constraint on every KashVector tool, not just a style choice. Beyond that, a few things are worth holding in mind before reading too much into any single number:

• A premium is partly the model's own caution. The underlying DCF caps growth assumptions hard by design. That structurally values fast-growing companies below where the market prices them — an across-the-board premium in a strong market is expected, not on its own evidence of mispricing.
• AI and platform-shift companies are the model's weak spot. A five-year cash-flow model can't price a genuine platform shift, and it penalises a company whose growth is currently showing up as capital expenditure rather than free cash flow. Semiconductor and cloud-infrastructure names read as the steepest premiums here partly for that reason.
• A theme can stay "expensive" for years while still compounding. A conservative intrinsic value is a reference point, not a ceiling — the market has traded well above conservative fair value for long stretches before, in both directions.
• Coverage matters. A fund ranked at 80% coverage is telling you about four-fifths of its weight, not all of it — the lower the coverage, the more caution the headline number deserves.
• It's a monthly snapshot, not a live feed. Holdings, prices and the underlying financial data are all frozen at the date shown — a fast-moving month can leave the numbers stale before the next refresh.