IPO backtest: methodology for IPOs from SEC prospectuses
The IPO backtest calculates what would have become of US IPOs since 2001 if you had filtered them using information available at entry, bought them, and held them for a fixed period. This methodology describes what the code behind the IPO backtest page does: where the list of IPOs comes from, how prices are matched, how entry, holding period, and comparison are calculated, and where the limitations lie.
List of IPOs from the prospectuses
The basis is the SEC full-text search. For each quarter since the first quarter of 2001, the nightly run searches all final prospectuses (Form 424B4) containing the phrase “initial public offering price”. From the cover page of each prospectus it reads the offer price per share, the number of shares offered (without the over-allotment), the ticker symbol, the exchange (Nasdaq, NYSE, NYSE American), the SEC industry code, the headquarters, and the state of incorporation. Some old filings from around 2001 exist only as a complete submission file; in that case it reads that file.
- IPO or follow-on offering: companies that are already listed also file prospectuses with this phrase. If the company had already filed annual or quarterly reports before the prospectus (its list of SEC filings), the prospectus only counts as an IPO if the cover page calls itself one (“this is our initial public offering”, “no public market”). For each company, the first IPO counts.
- Blank-check companies (SPACs): detected by industry code 6770, by units of a share and a warrant, or by their own description. They are excluded by default, because their price depends on the trust account until they acquire a company.
- ADS: shares of foreign companies in the form of depositary receipts are flagged; the filter “US-based companies only” excludes them via headquarters and state of incorporation.
- Offering size: offer price times shares offered according to the cover page.
A quarter counts as fully read once all hits have been read and the quarter is at least 45 days in the past (late filings). The nightly run reads for at most 20 minutes per night, newest quarters first; until all quarters are read, the page says how many are missing.
Prices from the first trading day
For each IPO, the nightly run loads the prices of the ticker and stores only derived values: returns for each entry and holding period, the exit date, the maximum drawdown, and a monthly path over five years. Returns include distributions (reinvested), before costs and taxes. Prices are also available for stocks that were later acquired, went bankrupt, or were delisted. This keeps the backtest free of survivorship bias, which flatters studies that only use stocks listed today.
- Matching the ticker: a ticker can later pass to another company. It only matches if the first price falls at most 7 days before and 30 days after the prospectus date. Otherwise the run tries the company's current tickers according to the SEC (renamed companies); if none matches, the IPO drops out with a reason and is retried after 90 days.
- Delisted: if the price series ends more than 10 days before the market's latest price, the stock counts as delisted. Holding periods that extend beyond the end finish at its last price: for an acquisition usually the offer, for a bankruptcy the last trading day.
- Not traded long enough yet: if a holding period has not been reached yet, the IPO is missing from exactly that holding period; the nightly run loads it again as soon as the next one is reached.
Entry, holding period, and comparison
Holding periods are counted in trading days: 20 and 60 days, one year (250), three years (750), and five years (1,250), always from entry. Entries:
- First close: available to anyone and the default.
- 5 or 20 trading days later: if you want to wait out the initial frenzy.
- Offer price: only possible with an allocation, which for popular IPOs is often not available at all or only partially. The return is then the first-day return times the return from the first close.
- raw first close versus the offer price (without later adjustments)
- return including distributions from the first close to exit
Comparison: the same period in the market (S&P 500) and in small caps (Russell 2000), each through an ETF with distributions, from the entry date to the exit date of each IPO. The excess return is the return minus the market return over the same period; the page shows the median and mean, the hit rate (share with a gain), and the share that beat the market.
The portfolio of all purchases approximates “buy every matching IPO”: at the end of each month, all IPOs currently held are equally weighted. Each IPO enters at the end of its first month and stays for about as many months as the holding period (trading days divided by 21). This is not a tradable strategy with costs, but it shows how a book of IPOs would have performed against the market over the years.
Filters using only information from entry
A backtest may only know what was known at entry; otherwise it looks into the future (look-ahead bias). That is why there are only filters using information from the prospectus or the first trading day:
- Period (year of the first trading day), sector by SEC industry code in twelve groups plus “other”, exchange, US headquarters, SPACs on or off.
- Minimum offer price ($5 by default, because smaller IPOs are often barely tradable) and offering size from and to in millions of dollars.
- First-day return: only known from the first close. With entry at the offer price it would be a look into the future; the page rejects this combination.
- Financials before the IPO: revenue, revenue growth, and net income (profitable or loss-making) for the last fiscal year that ended at least 30 days before the prospectus. They are in the prospectus; in machine-readable form they come from later annual reports (prior-year figures in the SEC's XBRL data, for each concept and calendar year for all companies). Growth is revenue versus the fiscal year before (300 to 430 days earlier). Figures are available from about 2009; older IPOs drop out under these filters, and the page says how many.
The result also splits the same IPOs into groups: thirds by offering size, offer price, first-day return, revenue, and growth, plus exchange, sector, headquarters, and pre-IPO profitability. Groups with fewer than five IPOs are left out.
What research shows about IPOs
Two findings shape the research. First, the first close is on average well above the offer price (underpricing): those who get an allocation earn money on the first day. Ritter and Welch (2002) summarize the explanations and see the main driver not in asymmetric information but in conflicts of interest and behavioral reasons. Second, IPOs often do worse over longer periods than comparable stocks: Ritter (1991) found this for the US over three years, Loughran and Ritter (1995) also over five years and for seasoned equity offerings of companies that were already listed. How strong the second finding is depends on the benchmark (market, companies of the same size, the same valuation) and on the weighting: equally weighted, the many small IPOs count more.
The IPO backtest therefore shows the median and mean side by side (a few big winners pull the mean above the median), compares with the market and with small caps, and lets you filter by size and financials. This lets you check for yourself whether the finding holds for a subset.
Limitations
- US only: IPOs on US exchanges with a prospectus filed with the SEC; IPOs only in Europe or Asia are missing.
- Ticker not found: IPOs without matching prices from the first day are missing (the page states how many). This tends to affect very small and very old IPOs.
- Cover page: price, number of shares, and ticker come from the text of the cover page. Unusual cover pages can lead to missing or wrong data; matching against the first price catches gross errors in the ticker, not in the size.
- Financials: taken from the most recently filed report for the year; a later restatement can differ from the prospectus. For each year the first available revenue concept counts (total revenues before revenue from contracts with customers).
- No trading: no costs, no taxes, no allocation rate. Especially with entry at the offer price, the result is an upper bound.
Frequently asked questions
Why is a well-known IPO missing?
Common reasons: the prospectus falls in a quarter the nightly run has not read yet; it was a direct listing without a prospectus with an offer price; the ticker had no prices from the first day; or the IPO did not take place on a US exchange.
Is the IPO backtest free of survivorship bias?
Yes, as far as the prices go: stocks that were later acquired, went bankrupt, or were delisted are included with their last price. Only IPOs with no matching prices at all are missing; their number is shown on the page.
Why can't I filter by first-day return when I enter at the offer price?
Because the first-day return is only known after the first trading day. If you buy at the offer price, you cannot yet know whether the first price will be higher. Such a filter would be a look into the future and would flatter the result.
What is the difference between median and mean?
The median is the middle IPO, the mean is the average. For IPOs they are often far apart, because a few very successful stocks pull the mean up while the typical IPO does worse.
Where do revenue and earnings before the IPO come from?
From the SEC's XBRL data. Prospectuses themselves are usually not machine-readable, but the figures for the last fiscal year before the IPO appear in the first annual report as prior-year figures. That is why they are only available from about 2009.
Literature
- Loughran, Tim; Ritter, Jay R. (1995): The New Issues Puzzle. The Journal of Finance 50 (1), pp. 23-51. doi:10.1111/j.1540-6261.1995.tb05166.x
- Ritter, Jay R. (1991): The Long-Run Performance of Initial Public Offerings. The Journal of Finance 46 (1), pp. 3-27. doi:10.1111/j.1540-6261.1991.tb03743.x
- Ritter, Jay R.; Welch, Ivo (2002): A Review of IPO Activity, Pricing, and Allocations. The Journal of Finance 57 (4), pp. 1795-1828. doi:10.1111/1540-6261.00478
This page describes how Portfolio Lab calculates; past results are not a forecast. All terms in the glossary.