
Second-Favourite Horses Lost 408 Units. Australia Accounted for Most of It.
The second favourite in a horse race occupies an awkward place in the betting market. It is not the obvious first choice, but it is not the outsider whose price promises an extraordinary payday. That middle position can feel attractive: perhaps the market has overestimated the favourite, while the next-best horse offers a more generous return. We put that idea under examination using 9,630 settled BACK bets from Trade Carefully's Global Second Favourite Horse Collector, recorded between 25 September and 17 November 2026. Only 2,024 selections won, a strike rate of 21.02%. On a one-unit stake per selection, with an assumed 2% commission on winning profits, the collector lost 407.89 units: a return of −4.24%. The most important finding, however, was not the overall loss. It was where the losses came from.
This is a Research Lab analysis of recorded small real-money selections, not a hypothetical backtest. Our Live Experiments programme explains the broader approach. Each winning BACK bet earns (matched decimal odds minus one) multiplied by 0.98 normalised units; every loser costs one unit. The figures below are comparable research units rather than the original cash profit and loss, and they do not establish a future betting advantage.

Australia carried nearly two-thirds of the damage
Across 3,273 Australian selections, only 646 won. The Australian subset lost 261.31 units, a return of −7.98%. That is 64% of the collector's total 407.89-unit loss. Removing Australia from the historical arithmetic leaves 6,357 bets and a loss of 146.58 units, approximately −2.31%. The exclusion would have reduced the deficit substantially, but it would not have transformed the original worldwide strategy into a winner. This distinction matters because it prevents a genuine research observation from becoming an exaggerated claim.
The contrast with our separate Favourite Horse Collector is particularly revealing. Australian favourites gained 72.53 normalised units across 3,344 eligible bets, while Australian second favourites lost 261.31 units. We therefore chose to retain Australia for the favourite strategy but exclude it from forward testing of second favourites. That was a response to a historical geographic split, not proof that the Australian market offers a permanent edge to one rank and a permanent disadvantage to another. The correct test begins after the rule is declared.
The United States told a very different story
The 3,127 US selections produced 732 winners and gained 72.37 units, a return of +2.31%. This was the only large country segment in clear profit. In the same workbook, Britain lost 78.62 units across 1,239 bets, France lost 48.59 across 666, Ireland lost 41.74 across 428 and New Zealand lost 47.27 across 547. South Africa was close to flat at −1.93 units from 342 selections. Those are not minor variations: they explain why treating the entire collector as one global method can conceal sharply different outcomes.
The US result is interesting enough to test, but it is not an invitation to announce a US-only winner. We discovered the country split by examining completed bets. A fresh, predeclared US-only sample would be needed to determine whether the historical +2.31% persists. Markets, race types, timing and selection mechanics may all differ across countries. The workbook provides the recorded country code, race description, selection, matched price and outcome; it does not prove why any country performed differently.
The opening night offered a miniature version of the experiment
On 25 September, the collector began with a run of US races. Go Grey at Churchill Downs, backed at 3.65, lost. Miz Clubcali at Santa Anita Park, priced at 4.80, lost. Oklahoma Smoke at Belmont Park at 3.60 lost. Magic Class at Churchill Downs at 7.20 lost, followed by Born To Rock at Santa Anita Park at 4.30 and Raging Bloom at Belmont Park at 5.30. Six selections, six lost units. The next race in that sequence, Big Time Coach at Santa Anita at 4.20, also lost. Those outcomes are recorded in the export; the finishing positions, pace and race incidents are not, so we should not invent a tale of bad luck at the final furlong.
Then came a win. Hes A Knockout at Santa Anita Park was backed at 5.00 and settled as a winner, earning 3.92 units after commission. Desert Kat, also at Santa Anita, won at 3.30 and earned 2.254 units. A bettor following the sequence could feel the balance shifting back. Yet even a win at 5.00 is not enough to pay for seven one-unit losses. The lesson is not that these particular horses were bad choices. It is that a strategy with roughly one winner in five needs prices and probabilities to align extremely well.
The losing months never produced a recovery
September's 1,053 bets lost 60.81 units (−5.77%). October added 5,768 selections and lost another 181.24 units (−3.14%). November's 2,809 bets through the 17th lost 165.84 units (−5.90%). The result was not driven by a single catastrophic afternoon. All three calendar months were negative. A large number of bets does not automatically make a collector reliable; it can simply make a consistently poor return harder to dismiss.
The pricing pattern adds another layer. The 4,911 selections at odds from 3.00 to below 5.00 lost 82.57 units (−1.68%). But the 4,086 selections from 5.00 to below 10.00 lost 317.25 units (−7.76%). That higher band contributed most of the overall damage. The 240 selections at odds of 10.00 or above lost 10.28 units (−4.28%), while 390 bets from 2.00 to below 3.00 earned a modest 2.10 units. Just three selections were below 2.00. These splits describe what happened, not which prices will work in future.
A spectacular win can disguise a much larger losing segment
The workbook contains memorable high-priced winners. On 14 October, Lexy was backed at Bathurst in Australia at 15.50 and won, earning 14.21 normalised units after commission. On 31 October, Sriracha won at Melton at 15.00, worth 13.72 units. Sweet Issy at Bunbury on 28 September won at 13.50, worth 12.25 units. Yet all three came from the Australian segment that lost 261.31 units overall. The individual wins were real, but they did not overturn the losses around them.
There is also a classification question worth retaining for further research. The Australian descriptions include events labelled ‘Pace M’, alongside conventional thoroughbred race descriptions. The export groups these under a single horse-racing collector, but harness racing and thoroughbred racing are not identical markets. Before attributing Australia's underperformance to geography alone, we should separate race disciplines and examine how the collector ranked runners in each. Otherwise, a country label may be standing in for a more important structural difference.
What the Australia exclusion actually achieves
Excluding Australia reduces the historical loss from 407.89 units to 146.58 units, but the remaining sample is still negative. Selecting the profitable US subset after the event would produce an even more flattering historical picture, yet that is precisely the kind of retrospective optimisation that can lead to false confidence. A defensible next experiment should record the Australia exclusion date, document whether harness and thoroughbred races are treated separately, and test the remaining selection rule without silently revising it whenever results disappoint.
The comparison with our Favourite Horse Collector is essential: Australian favourites and second favourites went in opposite directions, even though both came from the same wider research programme. Our
After 9,630 selections, the honest conclusion is that this worldwide second-favourite collector lost money, with Australia responsible for most of the damage and the US providing a genuine but unproven positive historical signal. The point of keeping the records is not to find a flattering way to retell the past. It is to decide which questions are worth asking next, and to make sure the next set of bets is capable of answering them.



Comments