WEBVTT

1
00:00:00.020 --> 00:00:00.200
Hello,

2
00:00:00.320 --> 00:00:02.402
welcome back to Papers with Backtest podcast.

3
00:00:02.803 --> 00:00:05.867
Today we dive into another algo trading research paper.

4
00:00:06.047 --> 00:00:06.387
Yes,

5
00:00:06.484 --> 00:00:08.484
this one's called Alpha Cloning Following

6
00:00:08.945 --> 00:00:09.547
13F Fillings.

7
00:00:10.133 --> 00:00:11.492
The authors are Randy Cohen,

8
00:00:11.773 --> 00:00:12.492
Christopher Polk,

9
00:00:12.672 --> 00:00:13.609
and Bernhard Sille.

10
00:00:14.016 --> 00:00:15.758
And the draft we're looking at is from

11
00:00:16.438 --> 00:00:17.359
March 18,

12
00:00:17.360 --> 00:00:18.438
2009.

13
00:00:18.439 --> 00:00:19.906
It's a really interesting premise,

14
00:00:19.938 --> 00:00:20.844
basically asking,

15
00:00:21.188 --> 00:00:24.422
can we look at what mutual fund managers report holding those

16
00:00:24.844 --> 00:00:28.109
13F filings and actually find profitable ideas?

17
00:00:28.375 --> 00:00:29.094
Specifically...

18
00:00:29.636 --> 00:00:31.538
They're like high conviction ideas.

19
00:00:31.557 --> 00:00:33.159
The paper calls them best ideas.

20
00:00:33.919 --> 00:00:36.423
Do those best ideas actually outperform the market?

21
00:00:36.684 --> 00:00:37.825
That's the core question.

22
00:00:37.942 --> 00:00:38.786
And for you listening,

23
00:00:39.184 --> 00:00:41.888
the potential payoff here is maybe a shortcut,

24
00:00:42.489 --> 00:00:44.770
a way to identify strategies that might generate alpha.

25
00:00:44.849 --> 00:00:45.130
Right.

26
00:00:45.450 --> 00:00:50.239
So our mission today is really to dig into the trading rules they tested and crucially the backtest results.

27
00:00:50.583 --> 00:00:52.692
How did these best ideas actually perform?

28
00:00:52.786 --> 00:00:52.895
OK,

29
00:00:53.145 --> 00:00:53.942
so let's start there.

30
00:00:54.336 --> 00:00:56.658
How did they even define a best idea?

31
00:00:56.659 --> 00:00:58.019
It can't just be the biggest holding,

32
00:00:58.080 --> 00:00:58.201
right?

33
00:00:58.202 --> 00:00:58.322
No,

34
00:00:58.361 --> 00:00:59.439
that's the key insight.

35
00:00:59.623 --> 00:01:04.509
They figured a manager's real conviction shows up when they overweight a stock compared to some baseline,

36
00:01:04.564 --> 00:01:05.244
some benchmark.

37
00:01:05.588 --> 00:01:06.228
Overweighting.

38
00:01:06.548 --> 00:01:08.853
So holding more of it than you'd kind of expect.

39
00:01:08.854 --> 00:01:09.369
Precisely.

40
00:01:09.572 --> 00:01:12.892
Like way more than its simple market cap might suggest,

41
00:01:12.893 --> 00:01:14.158
or more than other stocks they hold.

42
00:01:14.502 --> 00:01:16.064
It signals they really believe in it.

43
00:01:16.423 --> 00:01:16.892
Makes sense.

44
00:01:17.486 --> 00:01:19.455
And they didn't just use one benchmark.

45
00:01:19.642 --> 00:01:22.486
They had four different ways to measure this overweighting.

46
00:01:23.111 --> 00:01:23.220
or

47
00:01:23.392 --> 00:01:24.373
tilt as they call it.

48
00:01:24.414 --> 00:01:24.593
Yeah,

49
00:01:24.793 --> 00:01:26.074
four specific measures.

50
00:01:26.134 --> 00:01:27.816
The first one they called market tilt.

51
00:01:28.437 --> 00:01:28.636
Okay,

52
00:01:28.796 --> 00:01:29.417
market tilt.

53
00:01:29.597 --> 00:01:30.199
How does that work?

54
00:01:30.800 --> 00:01:31.523
It's pretty simple,

55
00:01:31.578 --> 00:01:31.921
actually.

56
00:01:32.140 --> 00:01:38.109
You just take the stock's weight in the fund's portfolio and subtract the stock's weight in the overall market portfolio.

57
00:01:38.226 --> 00:01:39.007
So market tilt,

58
00:01:39.109 --> 00:01:41.093
T-E-T equals I-T-I-M-P-O-O.

59
00:01:41.562 --> 00:01:45.281
A bigger number means more overweighting versus the market.

60
00:01:45.593 --> 00:01:45.937
Got it.

61
00:01:46.328 --> 00:01:48.156
Straightforward comparison to the whole market.

62
00:01:48.203 --> 00:01:49.453
What's the second tilt measure?

63
00:01:49.454 --> 00:01:51.312
The second is CAPM tilt.

64
00:01:51.776 --> 00:01:52.697
This one adds a layer.

65
00:01:52.857 --> 00:01:58.598
It takes that market tilt we just talked about and scales it by the stock's idiosyncratic variance.

66
00:01:58.676 --> 00:01:58.797
Whoa,

67
00:01:58.918 --> 00:01:59.039
OK.

68
00:01:59.578 --> 00:02:00.754
Idiosyncratic variance.

69
00:02:00.957 --> 00:02:01.137
Yeah,

70
00:02:01.317 --> 00:02:06.801
basically how much the stock's price moves around for reasons other than just following the market.

71
00:02:07.239 --> 00:02:10.535
They estimate it using like 60 days of return data.

72
00:02:10.973 --> 00:02:15.442
So the idea is if a manager overweights a stock that doesn't just track the market,

73
00:02:16.067 --> 00:02:18.160
that's an even stronger signal of conviction.

74
00:02:18.395 --> 00:02:19.301
That's the thinking.

75
00:02:19.800 --> 00:02:22.924
The formula is CAPM TILTY PSI.

76
00:02:23.363 --> 00:02:26.025
That sigma squared term is the idiosyncratic variance.

77
00:02:26.107 --> 00:02:26.287
Okay,

78
00:02:26.306 --> 00:02:27.369
that makes intuitive sense.

79
00:02:27.428 --> 00:02:28.170
A bolder bet.

80
00:02:28.209 --> 00:02:28.888
What were the other two?

81
00:02:29.107 --> 00:02:30.912
The third is Portfolio Tilt.

82
00:02:31.529 --> 00:02:31.795
Here,

83
00:02:31.935 --> 00:02:32.974
the benchmark changes.

84
00:02:33.092 --> 00:02:34.092
Instead of the whole market,

85
00:02:34.435 --> 00:02:41.099
you compare the stock's weight in the fund to its weight in evaluated portfolio of all the stocks that specific manager holds.

86
00:02:41.287 --> 00:02:41.474
Ah,

87
00:02:41.756 --> 00:02:43.396
so comparing it to their other choices.

88
00:02:43.490 --> 00:02:43.928
Exactly.

89
00:02:43.990 --> 00:02:45.084
It's Portfolio TILTY

90
00:02:45.459 --> 00:02:46.131
IVA FEE.

91
00:02:46.624 --> 00:02:49.868
Are they leaning into this stock even compared to their own usual picks?

92
00:02:50.087 --> 00:02:50.448
Interesting.

93
00:02:50.449 --> 00:02:51.149
And the fourth one,

94
00:02:51.911 --> 00:02:53.512
I'm guessing it combines ideas.

95
00:02:53.731 --> 00:02:54.192
You got it.

96
00:02:54.493 --> 00:02:55.930
CAPM portfolio tilt.

97
00:02:55.977 --> 00:02:57.118
It's that portfolio tilt.

98
00:02:57.454 --> 00:03:00.055
Scaled again by the stock's idiosyncratic variance.

99
00:03:00.540 --> 00:03:03.337
So key APM portfolio tilty tilt equals it

100
00:03:03.821 --> 00:03:04.665
IPAA fee.

101
00:03:04.821 --> 00:03:04.977
Right.

102
00:03:05.024 --> 00:03:09.087
Applying that independent movement factor to the comparison within their own holdings.

103
00:03:09.165 --> 00:03:09.399
Yeah.

104
00:03:09.962 --> 00:03:10.805
And for all four,

105
00:03:11.071 --> 00:03:14.899
a higher tilt value suggests stronger belief from the manager.

106
00:03:15.284 --> 00:03:15.464
Okay,

107
00:03:15.465 --> 00:03:18.707
so we have these four ways to spot potential best ideas.

108
00:03:19.508 --> 00:03:21.309
What data did they use to test this,

109
00:03:21.871 --> 00:03:23.012
to see if it actually worked?

110
00:03:23.375 --> 00:03:24.914
They used standard sources,

111
00:03:25.195 --> 00:03:25.672
pretty much.

112
00:03:25.992 --> 00:03:27.898
CRSP for the stock return data.

113
00:03:28.219 --> 00:03:28.461
Right,

114
00:03:28.516 --> 00:03:30.461
the usual suspect for academic work.

115
00:03:30.656 --> 00:03:33.320
And Thomson Reuters for the mutual fund holdings data,

116
00:03:33.461 --> 00:03:34.242
specifically the

117
00:03:34.695 --> 00:03:35.351
13F filings.

118
00:03:35.367 --> 00:03:36.226
And the time period.

119
00:03:36.351 --> 00:03:37.570
It was a decent chunk of time,

120
00:03:37.633 --> 00:03:40.648
January 1991 through December 2005.

121
00:03:40.649 --> 00:03:41.086
Okay,

122
00:03:41.148 --> 00:03:43.070
and they looked at specific types of funds.

123
00:03:43.344 --> 00:03:43.484
Yeah.

124
00:03:43.644 --> 00:03:43.865
U.S.

125
00:03:43.925 --> 00:03:44.986
domestic equity funds.

126
00:03:45.106 --> 00:03:46.508
They filtered them a bit,

127
00:03:46.887 --> 00:03:48.450
had to have at least 20 stocks,

128
00:03:48.610 --> 00:03:50.473
over $5 million in assets.

129
00:03:50.793 --> 00:03:51.434
And importantly,

130
00:03:51.489 --> 00:03:55.215
they excluded index funds and tax managed funds,

131
00:03:55.293 --> 00:03:56.278
wanted active managers.

132
00:03:56.379 --> 00:03:56.754
Makes sense.

133
00:03:57.317 --> 00:03:57.559
All right.

134
00:03:57.637 --> 00:03:58.278
Definitions,

135
00:03:58.801 --> 00:03:59.043
data.

136
00:03:59.496 --> 00:03:59.637
Yeah.

137
00:03:59.638 --> 00:04:00.418
Let's get to the results.

138
00:04:00.528 --> 00:04:03.387
What happened when they built a portfolio of these best ideas?

139
00:04:03.465 --> 00:04:03.903
Did it work?

140
00:04:04.309 --> 00:04:05.528
This is where it gets pretty compelling,

141
00:04:05.543 --> 00:04:05.871
I think.

142
00:04:06.121 --> 00:04:11.465
They created an equal weighted portfolio of all the identified best ideas from all the managers in their sample.

143
00:04:12.096 --> 00:04:35.613
rebalanced quarterly and they found positive average monthly excess returns that's return returns above the risk-free rate across all four tilt measures we're talking ranges like 1.26 percent to 1.88 percent per month on average okay that's substantial per month per month but of course the next question is risk right where they just loading up on risky stuff exactly my thought

144
00:04:35.700 --> 00:04:36.782
Did they adjust for risk?

145
00:04:36.841 --> 00:04:37.301
They did.

146
00:04:37.362 --> 00:04:41.125
They ran factor regressions using the standard Carhartt four-factor model,

147
00:04:41.266 --> 00:04:41.524
you know,

148
00:04:42.008 --> 00:04:43.711
market size value momentum.

149
00:04:43.946 --> 00:04:44.110
Uh-huh.

150
00:04:44.446 --> 00:04:45.907
The Fama French plus momentum.

151
00:04:45.908 --> 00:04:46.227
Right.

152
00:04:46.711 --> 00:04:48.469
Even after adjusting for those factors,

153
00:04:48.493 --> 00:04:50.555
the portfolio still showed positive alpha.

154
00:04:51.313 --> 00:04:52.383
Depending on the tilt measure,

155
00:04:52.384 --> 00:04:53.039
it was around,

156
00:04:53.071 --> 00:04:53.430
let's see,

157
00:04:53.431 --> 00:04:54.680
0.29%

158
00:04:54.681 --> 00:04:55.680
to 0.78%

159
00:04:55.681 --> 00:04:56.055
per month,

160
00:04:56.196 --> 00:04:58.336
statistically significant in most cases.

161
00:04:58.446 --> 00:05:01.868
So still outperforming even after accounting for those common risk factors.

162
00:05:01.869 --> 00:05:02.446
That's impressive.

163
00:05:02.571 --> 00:05:02.727
Yeah.

164
00:05:02.852 --> 00:05:04.289
They also used a six-factor model,

165
00:05:04.305 --> 00:05:04.571
didn't they?

166
00:05:04.832 --> 00:05:05.032
Yes,

167
00:05:05.052 --> 00:05:06.293
they added two more factors,

168
00:05:06.474 --> 00:05:08.055
one for idiosyncratic volatility,

169
00:05:08.115 --> 00:05:09.457
that's stock-specific risk again,

170
00:05:09.516 --> 00:05:10.899
and one for short-term reversal.

171
00:05:11.117 --> 00:05:12.641
And what happened with six factors?

172
00:05:12.797 --> 00:05:13.844
Did the alpha disappear?

173
00:05:14.366 --> 00:05:14.646
Actually,

174
00:05:14.746 --> 00:05:14.867
no.

175
00:05:15.287 --> 00:05:16.208
In many cases,

176
00:05:16.308 --> 00:05:20.791
the six factor alphas were even higher and more statistically significant,

177
00:05:21.213 --> 00:05:25.537
ranging from about 0.39 percent up to 1.15 percent per month.

178
00:05:25.998 --> 00:05:28.021
Higher alpha with more risk factors.

179
00:05:28.076 --> 00:05:28.842
How does that work?

180
00:05:29.318 --> 00:05:29.498
Well,

181
00:05:29.521 --> 00:05:34.685
it suggests the standard factors weren't fully capturing the risks or characteristics of these stocks.

182
00:05:35.045 --> 00:05:36.795
These added factors might be relevant.

183
00:05:36.889 --> 00:05:37.451
Interesting.

184
00:05:37.795 --> 00:05:41.076
And did any tilt measures stand out with the six factor model?

185
00:05:41.810 --> 00:05:42.029
Yeah,

186
00:05:42.107 --> 00:05:43.982
the two portfolio based tilts.

187
00:05:44.294 --> 00:05:46.876
portfolio tilt and CAPM portfolio tilt,

188
00:05:46.956 --> 00:05:49.237
the ones comparing a stock to the manager's own holdings.

189
00:05:49.698 --> 00:05:51.280
They generally show the strongest results,

190
00:05:51.523 --> 00:05:52.964
higher alphas and T-stats.

191
00:05:53.081 --> 00:05:53.206
OK,

192
00:05:53.339 --> 00:05:58.245
so maybe looking at conviction relative to their other picks is the most powerful signal.

193
00:05:58.605 --> 00:05:58.730
Now,

194
00:05:58.792 --> 00:06:00.730
what about these best fresh ideas?

195
00:06:00.933 --> 00:06:01.573
What was that about?

196
00:06:01.730 --> 00:06:01.917
Ah,

197
00:06:02.026 --> 00:06:02.292
yes,

198
00:06:02.589 --> 00:06:03.355
that was another layer.

199
00:06:03.730 --> 00:06:09.245
Best fresh ideas were defined as those best ideas where the fund manager had actually increased their position during the most recent quarter.

200
00:06:09.464 --> 00:06:11.073
So not just holding a big position,

201
00:06:11.136 --> 00:06:12.417
but actively buying more.

202
00:06:12.605 --> 00:06:13.167
Exactly.

203
00:06:13.230 --> 00:06:13.870
The thinking is...

204
00:06:13.998 --> 00:06:15.199
that's an even stronger,

205
00:06:15.560 --> 00:06:17.341
more current signal of conviction.

206
00:06:17.380 --> 00:06:18.962
And did that pan out in the results?

207
00:06:19.302 --> 00:06:19.865
It really did.

208
00:06:20.005 --> 00:06:20.384
Table 2,

209
00:06:20.505 --> 00:06:21.525
panel B shows this.

210
00:06:22.228 --> 00:06:23.626
The risk-adjusted returns,

211
00:06:23.767 --> 00:06:25.751
especially those six-factor alphas,

212
00:06:26.087 --> 00:06:28.916
were consistently better for the best fresh ideas.

213
00:06:29.470 --> 00:06:32.041
They ranged from 0.46%

214
00:06:32.353 --> 00:06:34.775
up to about 1.27%

215
00:06:34.900 --> 00:06:35.416
per month.

216
00:06:35.681 --> 00:06:38.775
So the act of buying recently adds another layer of predictive power.

217
00:06:38.884 --> 00:06:40.806
That's a key potential trading role right there.

218
00:06:40.918 --> 00:06:42.019
Definitely seems like it.

219
00:06:42.239 --> 00:06:45.342
Focus on the recent buys among the high conviction holdings.

220
00:06:45.542 --> 00:06:48.327
Did they look at like how high the conviction needed to be?

221
00:06:48.647 --> 00:06:51.569
Does it pay to focus only on the absolute top bets?

222
00:06:51.827 --> 00:06:53.671
They did analyze that in table five.

223
00:06:53.772 --> 00:06:55.733
They broke down performance by different thresholds,

224
00:06:55.734 --> 00:06:57.077
the top 100 percent of tilts,

225
00:06:57.139 --> 00:06:57.967
top 50 percent,

226
00:06:58.296 --> 00:07:00.311
and then really zooming in on just the top five percent.

227
00:07:00.312 --> 00:07:00.421
Wait,

228
00:07:00.422 --> 00:07:02.483
did the performance get stronger at the very top?

229
00:07:02.624 --> 00:07:02.921
Yes,

230
00:07:03.467 --> 00:07:04.639
significantly stronger.

231
00:07:05.077 --> 00:07:06.702
The outperformance wasn't just present.

232
00:07:06.703 --> 00:07:06.905
It was.

233
00:07:07.050 --> 00:07:09.814
amplified when focusing on the most extreme tilts.

234
00:07:09.872 --> 00:07:10.853
Any standout numbers there?

235
00:07:11.115 --> 00:07:11.236
Oh,

236
00:07:11.294 --> 00:07:11.454
yeah.

237
00:07:11.654 --> 00:07:12.536
The most striping one,

238
00:07:12.556 --> 00:07:12.857
I think,

239
00:07:12.896 --> 00:07:13.939
was for the top 5%

240
00:07:14.197 --> 00:07:15.962
of the CAPM portfolio tilts.

241
00:07:16.501 --> 00:07:16.783
Remember,

242
00:07:16.822 --> 00:07:20.501
that's high conviction relative to their own holdings adjusted for idiosyncratic risk.

243
00:07:21.087 --> 00:07:25.048
That portfolio generated a six-factor alpha of 1.88%

244
00:07:25.049 --> 00:07:25.423
per month.

245
00:07:25.689 --> 00:07:26.095
Get out.

246
00:07:26.423 --> 00:07:27.704
1.88%

247
00:07:27.829 --> 00:07:28.298
per month.

248
00:07:28.486 --> 00:07:29.267
That's huge.

249
00:07:29.345 --> 00:07:29.814
It is.

250
00:07:29.986 --> 00:07:30.533
Annualized,

251
00:07:30.534 --> 00:07:32.454
that's over 22%.

252
00:07:32.455 --> 00:07:34.033
Now obviously that's a back test,

253
00:07:34.454 --> 00:07:35.314
but it suggests that

254
00:07:35.614 --> 00:07:36.895
isolating those really,

255
00:07:36.975 --> 00:07:40.379
really high conviction bets could be extremely powerful.

256
00:07:40.758 --> 00:07:41.860
That feels like a major finding.

257
00:07:41.899 --> 00:07:42.840
Forget diversification.

258
00:07:42.879 --> 00:07:45.223
Maybe it's about finding those few killer ideas.

259
00:07:45.903 --> 00:07:48.989
They also tested a best minus rest strategy.

260
00:07:49.106 --> 00:07:49.950
What was the point of that?

261
00:07:50.028 --> 00:07:50.247
Right.

262
00:07:50.512 --> 00:07:51.051
Table seven.

263
00:07:51.247 --> 00:07:52.137
That was a neat test.

264
00:07:52.465 --> 00:07:56.309
The idea was to go long the manager's single best idea,

265
00:07:56.856 --> 00:07:58.575
identified by one of the tilt measures,

266
00:07:58.840 --> 00:08:01.356
and simultaneously short the rest of that manager's portfolio,

267
00:08:01.825 --> 00:08:02.747
weighted proportionally.

268
00:08:03.634 --> 00:08:07.018
So directly betting that the best idea outperforms their other holdings,

269
00:08:07.438 --> 00:08:11.260
trying to isolate the skill and picking that top stock from their general style.

270
00:08:11.323 --> 00:08:11.819
Exactly.

271
00:08:11.920 --> 00:08:13.623
Controls for manager-specific effects,

272
00:08:13.741 --> 00:08:14.342
sector bets,

273
00:08:14.405 --> 00:08:15.108
that sort of thing.

274
00:08:15.202 --> 00:08:15.983
And did that work?

275
00:08:16.069 --> 00:08:18.225
Did the best idea actually beat the rest?

276
00:08:18.366 --> 00:08:18.772
It did.

277
00:08:19.006 --> 00:08:25.733
They found statistically significant positive six-factor alphas for this strategy across all four tilt measures.

278
00:08:25.858 --> 00:08:28.733
Really reinforces the point that there's something special about those top picks,

279
00:08:28.764 --> 00:08:29.139
doesn't it?

280
00:08:29.405 --> 00:08:30.405
It certainly seems to.

281
00:08:30.498 --> 00:08:32.530
It's not just that the manager is good overall.

282
00:08:32.798 --> 00:08:36.100
Their best idea genuinely seems better than their other ideas.

283
00:08:36.221 --> 00:08:36.481
Okay,

284
00:08:36.503 --> 00:08:38.303
so the single best idea seems potent.

285
00:08:38.745 --> 00:08:39.764
What if they included,

286
00:08:39.885 --> 00:08:40.206
say,

287
00:08:40.870 --> 00:08:44.385
the top three or top five ideas instead of just the single best?

288
00:08:44.589 --> 00:08:45.292
Did they look at that?

289
00:08:45.354 --> 00:08:46.432
They did in Table 8.

290
00:08:46.448 --> 00:08:49.432
They expanded the long side of that best minus rest strategy.

291
00:08:49.729 --> 00:08:50.870
So long top three,

292
00:08:51.292 --> 00:08:51.901
short the rest.

293
00:08:52.292 --> 00:08:53.057
Long top five,

294
00:08:53.354 --> 00:08:53.885
short the rest.

295
00:08:53.995 --> 00:08:54.167
Yep.

296
00:08:54.620 --> 00:08:55.401
And interestingly,

297
00:08:55.464 --> 00:08:59.260
what they generally found was that as you included more stocks on the long side,

298
00:08:59.292 --> 00:09:01.182
going from top one to top three to top five,

299
00:09:01.638 --> 00:09:03.999
the resulting alpha tended to decrease.

300
00:09:04.261 --> 00:09:04.501
Ah,

301
00:09:04.761 --> 00:09:08.745
so diluting it with slightly less best ideas watered down the effect.

302
00:09:08.847 --> 00:09:09.386
Seems that way.

303
00:09:09.605 --> 00:09:14.769
It supports the idea that the conviction signal is strongest or maybe most valuable for that very top pick.

304
00:09:14.988 --> 00:09:15.628
Fascinating.

305
00:09:15.933 --> 00:09:16.089
Okay,

306
00:09:16.472 --> 00:09:18.261
moving beyond the core performance,

307
00:09:18.574 --> 00:09:21.074
they also looked at characteristics of the stocks themselves,

308
00:09:21.120 --> 00:09:21.277
right?

309
00:09:21.292 --> 00:09:21.870
Like liquidity.

310
00:09:22.042 --> 00:09:22.261
Yes.

311
00:09:22.605 --> 00:09:26.167
Table 9 looks at liquidity using the bid-ask spread as a proxy.

312
00:09:26.464 --> 00:09:30.370
They split the best ideas into high liquidity and low liquidity groups.

313
00:09:30.542 --> 00:09:31.261
And what did they find?

314
00:09:31.698 --> 00:09:33.878
Does liquidity matter for cloning these ideas?

315
00:09:34.238 --> 00:09:35.759
It seemed to matter quite a bit.

316
00:09:36.220 --> 00:09:38.478
The finding was that the less liquid,

317
00:09:38.861 --> 00:09:41.982
best ideas generated the vast majority of the alpha.

318
00:09:42.224 --> 00:09:42.458
Really?

319
00:09:42.857 --> 00:09:44.263
So the harder to trade ones?

320
00:09:44.404 --> 00:09:44.583
Yeah.

321
00:09:44.841 --> 00:09:45.544
The more liquid,

322
00:09:45.661 --> 00:09:46.365
best ideas,

323
00:09:46.419 --> 00:09:48.107
the ones easy for everyone to trade,

324
00:09:48.482 --> 00:09:49.826
showed much weaker results,

325
00:09:50.107 --> 00:09:51.357
sometimes even negative alpha.

326
00:09:51.404 --> 00:09:51.826
Huh.

327
00:09:52.419 --> 00:09:57.060
So maybe the edge comes from ideas where information isn't instantly priced in,

328
00:09:57.544 --> 00:09:59.388
or where it's costly for others to trade.

329
00:09:59.706 --> 00:10:01.469
That's a plausible interpretation.

330
00:10:01.751 --> 00:10:04.313
It suggests you might want to look beyond the big,

331
00:10:04.516 --> 00:10:04.954
obvious,

332
00:10:05.157 --> 00:10:06.958
highly liquid names if you're trying this.

333
00:10:07.219 --> 00:10:07.461
Okay,

334
00:10:07.661 --> 00:10:08.403
less liquid.

335
00:10:09.020 --> 00:10:10.387
What about popularity?

336
00:10:10.543 --> 00:10:15.036
Did it matter if lots of other managers also flagged the same stock as a best idea?

337
00:10:15.598 --> 00:10:16.499
That's in table 10.

338
00:10:16.819 --> 00:10:21.883
They measured popularity by basically summing up the tilt ranks for a stock across all the managers.

339
00:10:22.265 --> 00:10:25.805
A stock that many managers tilt heavily towards is popular.

340
00:10:25.984 --> 00:10:26.226
Right.

341
00:10:26.390 --> 00:10:26.930
And the result?

342
00:10:27.226 --> 00:10:29.070
Better to follow the crowd or go against it?

343
00:10:29.375 --> 00:10:31.094
The result suggested going against the crowd,

344
00:10:31.133 --> 00:10:31.453
actually.

345
00:10:31.609 --> 00:10:35.890
The vast majority of the abnormal returns came from the least popular best ideas.

346
00:10:36.203 --> 00:10:39.047
So the high conviction bets that aren't on everyone else's radar.

347
00:10:39.375 --> 00:10:39.922
Exactly.

348
00:10:40.047 --> 00:10:45.281
It hints that maybe the contrarian best ideas are the ones with the most potential alpha left in them.

349
00:10:45.710 --> 00:10:46.191
Interesting.

350
00:10:46.791 --> 00:10:50.856
So less liquid and less popular seems to be the sweet spot.

351
00:10:51.457 --> 00:10:53.238
Did they look at the funds themselves?

352
00:10:53.297 --> 00:10:53.457
Like,

353
00:10:53.477 --> 00:10:57.363
does it matter if the fund is huge or tiny or really concentrated?

354
00:10:57.418 --> 00:10:59.645
They did briefly touch on that in tables 11,

355
00:10:59.746 --> 00:11:00.668
12 and 13.

356
00:11:01.285 --> 00:11:04.402
They looked at fund concentration using the Herfindahl index,

357
00:11:04.746 --> 00:11:09.512
focus based on the number of holdings and just fund size assets under management.

358
00:11:09.527 --> 00:11:10.652
Any strong conclusions there?

359
00:11:11.186 --> 00:11:14.149
The results weren't always uniformly statistically significant,

360
00:11:14.230 --> 00:11:15.771
but there was definitely a trend.

361
00:11:15.970 --> 00:11:16.091
OK.

362
00:11:16.329 --> 00:11:16.931
What was the trend?

363
00:11:17.192 --> 00:11:22.337
It seemed that best ideas coming from smaller funds and funds that were more concentrated,

364
00:11:22.399 --> 00:11:23.962
holding fewer stocks overall,

365
00:11:24.321 --> 00:11:25.438
tended to perform better.

366
00:11:26.204 --> 00:11:26.884
Maybe smaller,

367
00:11:26.985 --> 00:11:32.079
more focused managers have a by handle on their picks or less pressure to over diversify.

368
00:11:32.392 --> 00:11:33.267
That could be part of it.

369
00:11:33.423 --> 00:11:39.173
They might be nimbler or perhaps their best ideas haven't been diluted quite as much by holding hundreds of other stocks.

370
00:11:39.688 --> 00:11:39.876
Okay.

371
00:11:40.038 --> 00:11:41.240
That's a lot of ground covered.

372
00:11:41.880 --> 00:11:42.822
If we boil it down,

373
00:11:43.462 --> 00:11:44.904
what are the main takeaways for you,

374
00:11:45.283 --> 00:11:45.802
the listener,

375
00:11:46.224 --> 00:11:48.447
maybe thinking about this alpha cloning idea?

376
00:11:48.830 --> 00:11:48.986
Well,

377
00:11:49.048 --> 00:11:49.345
first,

378
00:11:49.346 --> 00:11:53.251
it seems managers do have some ability to pick stocks that outperform,

379
00:11:53.330 --> 00:11:54.955
at least their top conviction ones.

380
00:11:55.290 --> 00:11:55.509
Right.

381
00:11:55.572 --> 00:12:00.275
The best ideas defined by that overweighting show real alpha.

382
00:12:00.744 --> 00:12:01.103
Second,

383
00:12:01.478 --> 00:12:02.994
focusing on best fresh ideas,

384
00:12:03.025 --> 00:12:04.603
the ones they recently bought more of,

385
00:12:05.009 --> 00:12:06.165
might give you an extra edge.

386
00:12:06.462 --> 00:12:06.650
Uh-huh.

387
00:12:06.962 --> 00:12:07.744
Recency matters.

388
00:12:08.025 --> 00:12:08.400
Third,

389
00:12:08.538 --> 00:12:14.284
But the less liquid and less popular best ideas seem to be where most of the outperformance is found.

390
00:12:14.804 --> 00:12:16.405
Don't just chase the obvious names.

391
00:12:16.569 --> 00:12:17.726
Go where it's less crowded,

392
00:12:17.765 --> 00:12:18.007
maybe.

393
00:12:18.288 --> 00:12:18.866
Seems like it.

394
00:12:19.351 --> 00:12:19.913
And fourth,

395
00:12:20.171 --> 00:12:23.374
there's maybe a slight advantage to looking at the picks from smaller,

396
00:12:23.655 --> 00:12:25.101
more concentrated funds.

397
00:12:25.491 --> 00:12:32.819
And the big picture implication seems to be that maybe the reason many funds underperform overall is because they diversify too much,

398
00:12:33.397 --> 00:12:35.960
watering down the impact of their genuinely good ideas.

399
00:12:36.101 --> 00:12:37.913
That's certainly what the paper suggests.

400
00:12:37.998 --> 00:12:40.861
Their best insights get lost in a sea of other holdings.

401
00:12:41.082 --> 00:12:43.142
So lots of potential rules and ideas there.

402
00:12:43.404 --> 00:12:44.443
It definitely leaves you thinking.

403
00:12:44.763 --> 00:12:45.404
It really does.

404
00:12:45.826 --> 00:12:48.087
And it poses a provocative question for you listening.

405
00:12:48.689 --> 00:12:49.830
If this research holds up,

406
00:12:50.330 --> 00:12:52.189
how could you actually use public

407
00:12:52.533 --> 00:12:54.822
13F filings to implement some version of this?

408
00:12:55.181 --> 00:12:56.337
What are the practical hurdles,

409
00:12:56.338 --> 00:12:57.040
the data needs,

410
00:12:57.056 --> 00:13:00.103
the timing issues you'd face trying to become an alpha cloner?

411
00:13:00.900 --> 00:13:01.025
Yeah,

412
00:13:01.165 --> 00:13:04.337
bridging that gap from research paper to real-world trading strategy.

413
00:13:04.447 --> 00:13:05.384
That's the challenge.

414
00:13:05.712 --> 00:13:06.603
A great thought to end on.

415
00:13:06.918 --> 00:13:09.063
Thank you for tuning in to Papers with Backtest podcast.

416
00:13:09.305 --> 00:13:11.389
We hope today's episode gave you useful insights.

417
00:13:11.611 --> 00:13:13.635
Join us next time as we break down more research.

418
00:13:13.975 --> 00:13:15.361
And for more papers and backtests,

419
00:13:15.557 --> 00:13:18.205
find us at https.paperswithbacktest.com.

420
00:13:18.471 --> 00:13:19.025
Happy trading.