Societe Generale's trend model was down 2.0% on bonds at the end of June and up 5.6% by September 30, with 4.4 points of that earned in September alone. The gain is real and narrow: it came from US and Japanese shorts, not the Bund, and since June the bond sleeve is the only reason the model's book is not lower.
On June 30, the bond positions in Societe Generale's trend model were under water. Its daily trend report showed the bonds sleeve down 2.02% for the year, gross of fees. At midyear, the bond trade now credited for the quant windfall had yet to pay off.
It paid mostly in one month. By August 31 the sleeve was up 1.19%. On September 30 it stood at 5.57%, which puts 4.38 points, more than half of the 7.6 points the sleeve has gained since June, in September. One session on October 1 has already erased part of that gain, and another asset class accounts for more of the year's return. The popular story adds a flight to German Bunds, and the model's own page shows what that meant for a short.
September's gain came from a repricing of bonds. On the Treasury's daily par curve the 10-year yield was 4.79% on September 1 and 5.29% on September 30, while the 2-year went from 4.39% to 4.88%. A 10-year Treasury with a duration near 8 loses about 4% of its price when its yield rises half a point, and a short position collects roughly that. The trade was one large move in yields, held correctly.
The positions were not new. The model had been short 10-year Treasury futures for 195 days by October 1 and the 2-year for 303, and it sat through the losses of the first half before September paid for the patience. That is how trend following is meant to work, and why its returns look so lumpy on a chart.
Then the move paused. Reuters reported that the 10-year touched 5.34% on Thursday, its highest since 2002, after the biggest quarterly rise in yields this century, almost 90 basis points in the third quarter. That was an intraday peak. The Treasury's own par yield at the close on October 1 was 5.24%, five basis points below the day before, and for a model short the whole curve that was enough. On October 1 the bonds sleeve lost 0.53 points, about an eighth of September's gain in one session, and the whole book lost 0.93%. By Friday the 10-year had backed up to 5.28%, which shows how little room there is between a good day and a bad one.
The same table shows where the money came from. The short in Japanese government bonds, held 440 days, has earned 1.64% this year and the US 2-year short 1.57%. The Bund short is down 0.65%. Those returns match the market's movements. The FT reported on Friday that German debt had been the rout's pre-eminent haven, with the 10-year yield down about 0.17 points on the week while US, French and Italian yields rose. The Bund did touch a 17-year high in yield of about 3.65% earlier in the week, then settled near 3.52% on Thursday and eased to 3.46% by 15:58 CET on Friday. Over September as a whole it rose only about 0.11 points, against 0.50 for the US 10-year. A short on the German curve is a bet on the part of the market that held up best.
The rest of the book explains why bonds are not the whole story. Commodities were up 17.41% for the year at the end of September and 17.58% a day later, the largest contribution in the model, and the oldest of those longs are energy: gasoline futures held for 240 days and heating oil for 241, up 4.74% and 5.83% for the year. Currencies were down 8.98% and equity indices up 1.43%. Since June 30 the picture is sharper: bonds added 7.6 points while equity indices gave back about 6 and currencies about 5, which is why the model's gross book, 17.63% at the end of September, stands below the 19.11% of midyear.
Managers have nevertheless done well. The SG CTA Index, the managers' benchmark, is up 15.66% for the year, including 4.10% in September, against about 13% for the S&P 500 on a total return basis. That is a strong year. The figures also show where the return came from and how quickly it can reverse: Treasuries and JGBs in one month, commodities across the year, and a quick reversal when yields eased. Most trend models also size positions by volatility, so the spike in yields that made the money also pushes positions smaller, which is one reason a month like September is hard to repeat.
The model is not close to reversing its shorts. SocGen's watch list of positions nearest to flipping on October 1 contains no bond contract at all; it is equity indices, two currencies, coffee and cotton. The bond shorts have room before the model reverses them, which is the case for the trade and the reason a squeeze would hurt. The profit sits in the US curve and in Japan, so the sleeve's next move depends on Treasury yields staying above where September began, not on anything the Bund does.
The next move depends on the policy calendar. The Federal Reserve meets on October 27 and 28. A 10-year Treasury close above the September 30 level of 5.29% would extend the gain, and a return to the September 1 level of 4.79% would hand back the whole of September. SocGen's October figures, due in the first days of November, will show the bonds sleeve flat to negative, with the year's gains still carried by commodities.