WEBVTT

1
00:00:00.060 --> 00:00:00.260
Hello,

2
00:00:00.400 --> 00:00:02.643
welcome back to Papers with Backtest podcast.

3
00:00:02.963 --> 00:00:06.164
Today we dive into another algo trading research paper.

4
00:00:06.387 --> 00:00:09.367
And today we're looking at something called the accruals anomaly.

5
00:00:09.906 --> 00:00:10.094
Yeah,

6
00:00:10.211 --> 00:00:11.273
it's a really interesting one.

7
00:00:11.570 --> 00:00:12.094
Basically,

8
00:00:12.836 --> 00:00:17.312
it's this observation that companies reporting higher non-cash earnings,

9
00:00:18.281 --> 00:00:19.641
those are the accounting accruals.

10
00:00:19.938 --> 00:00:22.234
They tend to have lower stock returns later on.

11
00:00:22.406 --> 00:00:22.656
Right.

12
00:00:22.750 --> 00:00:25.078
And the opposite to lower accruals,

13
00:00:25.312 --> 00:00:26.688
potentially better returns.

14
00:00:27.500 --> 00:00:28.312
It seems a bit.

15
00:00:28.384 --> 00:00:28.824
it backward,

16
00:00:28.884 --> 00:00:29.265
doesn't it?

17
00:00:29.345 --> 00:00:29.946
It really does.

18
00:00:29.986 --> 00:00:30.767
You'd normally think,

19
00:00:30.786 --> 00:00:31.046
you know,

20
00:00:31.228 --> 00:00:33.728
higher reported earnings equals good news.

21
00:00:33.849 --> 00:00:34.310
Exactly.

22
00:00:34.751 --> 00:00:38.415
But this suggests maybe we need to look closer at how those earnings are generated,

23
00:00:38.736 --> 00:00:40.376
especially those accounting adjustments.

24
00:00:40.579 --> 00:00:42.962
And the paper we're unpacking today really gets into this.

25
00:00:43.017 --> 00:00:49.243
It's The Persistence of the Accruals Anomaly by Baruch Lev and Dora Nesim from back in April 2004.

26
00:00:49.603 --> 00:00:50.618
And the fascinating thing,

27
00:00:50.665 --> 00:00:51.087
like you said,

28
00:00:51.088 --> 00:00:51.884
is persistence.

29
00:00:52.134 --> 00:00:53.853
This wasn't a new discovery even then.

30
00:00:53.915 --> 00:00:55.087
People had known about it for.

31
00:00:55.376 --> 00:00:55.496
what,

32
00:00:55.556 --> 00:00:56.237
almost a decade?

33
00:00:56.277 --> 00:00:56.457
Yeah,

34
00:00:56.537 --> 00:00:57.317
nearly 10 years.

35
00:00:57.318 --> 00:01:00.843
And the basic trading strategy seems pretty straightforward on paper.

36
00:01:01.183 --> 00:01:03.265
You mean buy the stocks with low accruals,

37
00:01:03.562 --> 00:01:05.124
bet against the ones with high accruals?

38
00:01:05.288 --> 00:01:05.905
Precisely.

39
00:01:06.608 --> 00:01:07.569
So the big question,

40
00:01:07.710 --> 00:01:09.491
the one we're really tackling for you today,

41
00:01:10.374 --> 00:01:12.437
is why didn't this just disappear?

42
00:01:12.687 --> 00:01:12.874
Right.

43
00:01:13.140 --> 00:01:14.671
Shouldn't sophisticated investors,

44
00:01:14.718 --> 00:01:15.155
you know,

45
00:01:15.156 --> 00:01:15.874
the big funds,

46
00:01:16.218 --> 00:01:17.765
have arbitraged this away by now?

47
00:01:17.921 --> 00:01:18.718
That's the puzzle.

48
00:01:19.202 --> 00:01:23.108
And what can we actually take away from this for trading rules or backtesting?

49
00:01:23.483 --> 00:01:23.671
Well...

50
00:01:23.952 --> 00:01:25.513
If something like this hangs around for so long,

51
00:01:25.532 --> 00:01:26.733
there must be some friction,

52
00:01:27.192 --> 00:01:28.595
some reason it's hard to capture.

53
00:01:29.294 --> 00:01:30.833
Let's see what Lev and Nassim found.

54
00:01:31.095 --> 00:01:31.313
Okay,

55
00:01:31.395 --> 00:01:32.395
so first things first.

56
00:01:33.153 --> 00:01:35.513
Did the anomaly actually exist consistently?

57
00:01:35.552 --> 00:01:37.052
They looked at a really long period,

58
00:01:37.099 --> 00:01:37.294
right?

59
00:01:37.295 --> 00:01:37.372
Yeah,

60
00:01:37.373 --> 00:01:39.817
1965 all the way to 2002.

61
00:01:40.122 --> 00:01:41.669
A pretty solid chunk of time.

62
00:01:41.794 --> 00:01:44.934
And they calculated these accruals using a couple of different methods,

63
00:01:44.981 --> 00:01:45.278
I think.

64
00:01:45.481 --> 00:01:45.903
That's right.

65
00:01:46.309 --> 00:01:47.716
One was based on the balance sheet data.

66
00:01:47.731 --> 00:01:49.122
They called it BSACCC.

67
00:01:49.731 --> 00:01:52.841
It looks at changes in things like current assets minus cash.

68
00:01:53.580 --> 00:01:55.202
current liabilities minus debt.

69
00:01:55.442 --> 00:01:55.582
Yeah.

70
00:01:55.663 --> 00:01:55.903
You know,

71
00:01:56.143 --> 00:01:58.606
trying to isolate that non-cash part of earnings.

72
00:01:58.625 --> 00:01:58.747
OK,

73
00:01:58.887 --> 00:01:59.184
got it.

74
00:01:59.344 --> 00:01:59.766
And the other?

75
00:01:59.985 --> 00:02:01.848
The other used the cash flow statement,

76
00:02:02.028 --> 00:02:02.832
CFSACC.

77
00:02:03.426 --> 00:02:04.614
So two angles on the same idea.

78
00:02:04.989 --> 00:02:08.293
So what did they find when they actually tested the strategy,

79
00:02:08.918 --> 00:02:09.801
the long short one?

80
00:02:09.895 --> 00:02:10.020
Well,

81
00:02:10.021 --> 00:02:10.723
they confirmed it.

82
00:02:10.754 --> 00:02:13.051
The anomaly was definitely there over that whole period.

83
00:02:13.145 --> 00:02:14.098
How did they test it?

84
00:02:14.161 --> 00:02:15.145
Like a portfolio?

85
00:02:15.348 --> 00:02:15.457
Yeah,

86
00:02:15.489 --> 00:02:17.676
they set up what's called a zero investment portfolio.

87
00:02:18.286 --> 00:02:18.692
Basically,

88
00:02:18.723 --> 00:02:21.848
you go long the bottom 10 percent of stocks,

89
00:02:21.849 --> 00:02:23.192
the ones with the lowest accruals.

90
00:02:23.480 --> 00:02:23.881
And short,

91
00:02:23.901 --> 00:02:24.681
the top 10%,

92
00:02:24.702 --> 00:02:25.381
the highest accrual.

93
00:02:25.462 --> 00:02:25.862
Exactly.

94
00:02:26.063 --> 00:02:26.803
And the returns.

95
00:02:27.204 --> 00:02:29.946
They were statistically significant and positive.

96
00:02:30.465 --> 00:02:31.891
How positive are we talking?

97
00:02:32.649 --> 00:02:34.954
Depending on the exact measure and sample,

98
00:02:35.313 --> 00:02:39.469
the average abnormal returns were in the range of about 7.5%

99
00:02:39.516 --> 00:02:40.860
to 8.9%

100
00:02:40.861 --> 00:02:41.251
per year.

101
00:02:41.641 --> 00:02:42.673
That's from their table one.

102
00:02:43.001 --> 00:02:43.219
Okay.

103
00:02:43.985 --> 00:02:44.719
That's not trivial.

104
00:02:45.094 --> 00:02:46.360
But here's the kicker you mentioned.

105
00:02:46.438 --> 00:02:46.688
Right.

106
00:02:46.782 --> 00:02:47.469
The magnitude,

107
00:02:47.485 --> 00:02:48.548
the strength of this effect.

108
00:02:49.907 --> 00:02:51.735
It hadn't really diminished much over time.

109
00:02:51.832 --> 00:02:53.372
So even though people knew about it,

110
00:02:53.692 --> 00:02:55.493
the potential profit was still there?

111
00:02:55.554 --> 00:02:56.015
Pretty much.

112
00:02:56.675 --> 00:02:57.335
Their analysis,

113
00:02:57.433 --> 00:02:58.413
looking for a trend,

114
00:02:58.815 --> 00:03:00.456
didn't show a significant decrease.

115
00:03:00.956 --> 00:03:02.815
That's the core puzzle we keep coming back to.

116
00:03:03.112 --> 00:03:05.292
It really makes you wonder about the big players then,

117
00:03:05.495 --> 00:03:06.651
the institutional investors.

118
00:03:06.894 --> 00:03:07.714
What were they doing?

119
00:03:08.073 --> 00:03:08.214
Well,

120
00:03:08.355 --> 00:03:09.292
the paper looked into that.

121
00:03:09.730 --> 00:03:12.839
They examined institutional ownership changes in these companies.

122
00:03:12.901 --> 00:03:13.683
Did they react?

123
00:03:13.933 --> 00:03:14.917
Were they jumping on this?

124
00:03:15.355 --> 00:03:16.073
They did react,

125
00:03:16.151 --> 00:03:16.417
yes.

126
00:03:17.073 --> 00:03:21.151
The data showed institutions were trading based on Krul's info.

127
00:03:22.012 --> 00:03:25.155
Mostly in the first couple of quarters after the fiscal year ends,

128
00:03:25.215 --> 00:03:27.758
which is when you'd expect that information to be digested.

129
00:03:28.118 --> 00:03:28.239
OK,

130
00:03:28.360 --> 00:03:29.938
so they weren't completely ignoring it.

131
00:03:30.243 --> 00:03:30.360
No,

132
00:03:30.641 --> 00:03:31.040
not at all.

133
00:03:31.422 --> 00:03:32.282
And interestingly,

134
00:03:33.344 --> 00:03:35.665
it seemed like the more active institutions,

135
00:03:35.727 --> 00:03:37.961
the ones they call transients who trade more often,

136
00:03:38.805 --> 00:03:40.946
they accounted for a big chunk of this reaction.

137
00:03:41.196 --> 00:03:43.571
Even if they didn't hold the biggest overall position.

138
00:03:43.711 --> 00:03:43.946
Right.

139
00:03:44.055 --> 00:03:44.868
But here's the thing.

140
00:03:45.743 --> 00:03:46.821
Even with this reaction,

141
00:03:46.915 --> 00:03:48.165
the anomaly persisted.

142
00:03:48.446 --> 00:03:50.540
So their trading just wasn't enough to like.

143
00:03:50.820 --> 00:03:51.461
close the gap?

144
00:03:51.521 --> 00:03:52.241
Apparently not.

145
00:03:52.762 --> 00:03:57.366
It seems the institutional reaction was just too weak overall to make the anomaly disappear.

146
00:03:57.608 --> 00:03:57.749
Hmm.

147
00:03:58.206 --> 00:03:58.589
So why?

148
00:03:58.667 --> 00:04:00.046
Why wasn't the response stronger?

149
00:04:00.108 --> 00:04:01.632
Did the paper offer reasons?

150
00:04:02.014 --> 00:04:02.374
It did.

151
00:04:02.694 --> 00:04:05.194
It looked at the characteristics of the companies at the extremes,

152
00:04:05.514 --> 00:04:08.014
the really high and really low accrual firms.

153
00:04:08.155 --> 00:04:09.124
And what did they find?

154
00:04:09.171 --> 00:04:10.280
What are these companies like?

155
00:04:10.546 --> 00:04:10.686
Well,

156
00:04:10.687 --> 00:04:14.577
they tend to have traits that institutions often try to avoid.

157
00:04:14.921 --> 00:04:15.421
Like what?

158
00:04:15.530 --> 00:04:16.592
Like being small companies.

159
00:04:16.639 --> 00:04:18.046
There was a negative correlation there.

160
00:04:18.186 --> 00:04:18.374
Okay.

161
00:04:18.483 --> 00:04:19.327
Smaller market cap.

162
00:04:19.604 --> 00:04:20.025
Makes sense.

163
00:04:20.085 --> 00:04:21.386
Harder for big funds to trade.

164
00:04:21.726 --> 00:04:22.107
What else?

165
00:04:22.767 --> 00:04:23.527
Lower stock prices,

166
00:04:23.808 --> 00:04:24.027
often.

167
00:04:24.507 --> 00:04:24.730
Also,

168
00:04:24.890 --> 00:04:26.472
a lower book-to-market ratio.

169
00:04:26.570 --> 00:04:26.691
Now,

170
00:04:26.750 --> 00:04:30.296
institutions often prefer higher book-to-market sort of value stocks,

171
00:04:30.718 --> 00:04:34.218
but these extreme accrual firms tended to be lower on that scale.

172
00:04:34.336 --> 00:04:34.562
Right.

173
00:04:34.601 --> 00:04:37.406
Liquidity and maybe style preferences playing a role.

174
00:04:37.484 --> 00:04:37.968
Exactly.

175
00:04:38.062 --> 00:04:38.484
And also,

176
00:04:39.000 --> 00:04:40.296
higher residual volatility,

177
00:04:40.390 --> 00:04:40.765
basically,

178
00:04:40.843 --> 00:04:42.171
more stock-specific risk,

179
00:04:42.625 --> 00:04:43.890
and lower profitability.

180
00:04:44.203 --> 00:04:44.328
So,

181
00:04:44.531 --> 00:04:44.953
smaller,

182
00:04:45.046 --> 00:04:45.734
lower price,

183
00:04:45.796 --> 00:04:46.187
maybe more

184
00:04:46.328 --> 00:04:49.809
growth oriented by the Bia member riskier and less profitable.

185
00:04:49.970 --> 00:04:50.108
Yeah.

186
00:04:50.149 --> 00:04:54.468
A whole cluster of characteristics that aren't typically on the institutional favorites list.

187
00:04:54.647 --> 00:04:56.030
Do they confirm this statistically?

188
00:04:56.171 --> 00:04:56.327
Yeah.

189
00:04:56.671 --> 00:05:00.233
Table four in the paper shows regressions confirming these relationships.

190
00:05:00.851 --> 00:05:02.944
Institutions generally prefer the bigger firms,

191
00:05:03.179 --> 00:05:03.944
higher share prices,

192
00:05:04.007 --> 00:05:04.132
etc.

193
00:05:04.778 --> 00:05:10.623
So it seems the very stocks where this anomaly is strongest are the ones the big players tend to shy away from anyway.

194
00:05:11.026 --> 00:05:12.467
That seems to be a big part of the story.

195
00:05:13.186 --> 00:05:14.006
Their mandates,

196
00:05:14.264 --> 00:05:15.326
their risk controls,

197
00:05:15.709 --> 00:05:20.756
their preferences just don't align well with heavily trading these specific types of stocks.

198
00:05:21.381 --> 00:05:21.490
OK,

199
00:05:21.647 --> 00:05:25.084
so if the institutions aren't fully stepping in because they don't like the stocks,

200
00:05:25.194 --> 00:05:26.865
what about individual investors?

201
00:05:26.975 --> 00:05:28.053
Couldn't they capture this?

202
00:05:29.147 --> 00:05:29.256
Well,

203
00:05:29.287 --> 00:05:30.506
that brings us to the next hurdle,

204
00:05:31.287 --> 00:05:32.381
transaction costs.

205
00:05:32.878 --> 00:05:33.098
Right.

206
00:05:33.118 --> 00:05:36.141
The practical side of actually putting the trades on?

207
00:05:36.323 --> 00:05:40.365
The paper argues that trading these extreme accruals firms,

208
00:05:40.467 --> 00:05:41.623
especially for individuals,

209
00:05:42.045 --> 00:05:43.967
involves pretty substantial costs.

210
00:05:44.272 --> 00:05:46.053
Why would they be particularly high here?

211
00:05:46.553 --> 00:05:46.694
Well,

212
00:05:46.695 --> 00:05:47.389
a couple of reasons.

213
00:05:47.592 --> 00:05:47.873
First,

214
00:05:48.194 --> 00:05:50.990
their simulations suggested that to get reliable,

215
00:05:51.162 --> 00:05:54.397
statistically significant positive returns from this strategy,

216
00:05:55.225 --> 00:05:56.897
you couldn't just pick one or two stocks.

217
00:05:57.178 --> 00:05:58.381
You needed diversification.

218
00:05:58.740 --> 00:05:58.990
Yes,

219
00:05:59.209 --> 00:05:59.647
quite a bit.

220
00:06:00.278 --> 00:06:05.803
They estimated you'd need around 40 securities in the long portfolio and another 40 in the short portfolio.

221
00:06:06.022 --> 00:06:06.284
Wow.

222
00:06:06.546 --> 00:06:07.667
80 stocks total.

223
00:06:07.706 --> 00:06:08.846
That's a lot of trades.

224
00:06:09.143 --> 00:06:09.448
It is.

225
00:06:09.745 --> 00:06:11.667
And each trade has commissions,

226
00:06:11.831 --> 00:06:12.667
potential slippage,

227
00:06:13.409 --> 00:06:14.471
especially with smaller,

228
00:06:14.815 --> 00:06:16.112
maybe less liquid stocks.

229
00:06:16.651 --> 00:06:21.245
The fixed costs per trade really start to add up when you're trading that many names.

230
00:06:21.839 --> 00:06:23.698
I can see how that would eat into profits quickly,

231
00:06:23.760 --> 00:06:25.104
especially for a smaller account.

232
00:06:25.229 --> 00:06:25.620
Definitely.

233
00:06:26.135 --> 00:06:27.510
And maybe even more importantly,

234
00:06:27.901 --> 00:06:29.885
remember where a lot of the profit comes from.

235
00:06:30.274 --> 00:06:30.975
The short side,

236
00:06:31.295 --> 00:06:33.318
betting against the high accrual companies.

237
00:06:33.456 --> 00:06:33.958
Exactly.

238
00:06:34.300 --> 00:06:36.378
And short selling has its own specific,

239
00:06:36.620 --> 00:06:37.960
often significant costs.

240
00:06:38.081 --> 00:06:39.222
Like borrowing fees.

241
00:06:39.784 --> 00:06:40.245
Precisely.

242
00:06:40.581 --> 00:06:42.628
You have to pay to borrow the stock you want to short.

243
00:06:43.167 --> 00:06:44.511
And those fees can vary a lot.

244
00:06:44.886 --> 00:06:46.667
The paper cited estimates anywhere from

245
00:06:48.058 --> 00:06:50.808
0.20% up to nearly 4.72%

246
00:06:50.809 --> 00:06:51.151
per year.

247
00:06:51.495 --> 00:06:51.730
Ouch.

248
00:06:51.980 --> 00:06:56.261
That top end could wipe out a big chunk of the potential anomaly return right there.

249
00:06:56.433 --> 00:06:57.042
It really could.

250
00:06:57.394 --> 00:06:57.674
Plus,

251
00:06:57.814 --> 00:07:00.397
there's the risk of the stock being recalled by the lender,

252
00:07:00.858 --> 00:07:03.702
forcing you to close your position perhaps at an inconvenient time.

253
00:07:03.838 --> 00:07:04.920
So the shorting aspect,

254
00:07:04.940 --> 00:07:07.323
which seems crucial for the strategy's overall return,

255
00:07:07.502 --> 00:07:10.549
is particularly expensive and may be difficult for individuals.

256
00:07:10.963 --> 00:07:11.971
That's a key takeaway.

257
00:07:12.331 --> 00:07:13.588
High transaction costs,

258
00:07:14.198 --> 00:07:16.854
amplified by the need for broad diversification,

259
00:07:17.401 --> 00:07:22.104
and especially the high costs and complexities of shorting those high-curral stocks.

260
00:07:22.495 --> 00:07:24.557
So let's try and summarize this for everyone listening.

261
00:07:25.206 --> 00:07:26.207
The accruals anomaly,

262
00:07:26.347 --> 00:07:29.853
this negative link between non-cash earnings and future returns,

263
00:07:30.032 --> 00:07:30.993
it's been persistent.

264
00:07:31.290 --> 00:07:31.571
Right.

265
00:07:31.833 --> 00:07:35.274
And it seems to stick around likely because of a combination of factors.

266
00:07:35.517 --> 00:07:35.876
First,

267
00:07:36.017 --> 00:07:37.579
the big institutional players,

268
00:07:37.681 --> 00:07:38.618
while aware of it,

269
00:07:38.798 --> 00:07:41.657
tend to avoid the specific types of stocks small,

270
00:07:42.048 --> 00:07:42.767
low priced,

271
00:07:43.142 --> 00:07:43.798
volatile,

272
00:07:43.845 --> 00:07:45.220
where the anomaly is strongest.

273
00:07:45.704 --> 00:07:48.876
Their hands are kind of tied by their investment styles or mandates.

274
00:07:48.954 --> 00:07:49.126
Yeah,

275
00:07:49.204 --> 00:07:50.813
there's an institutional preference issue.

276
00:07:51.548 --> 00:07:52.126
And second,

277
00:07:52.579 --> 00:07:54.157
for individual investors who...

278
00:07:54.506 --> 00:07:56.248
might be more willing to trade these stocks.

279
00:07:56.569 --> 00:07:58.270
The transaction costs get in the way,

280
00:07:58.969 --> 00:08:05.074
especially the need to hold lots of stocks and the significant costs associated with short-selling the high accruals firms.

281
00:08:05.176 --> 00:08:05.637
Exactly.

282
00:08:05.856 --> 00:08:09.699
The costs of actually implementing the strategy seem to be a major barrier.

283
00:08:09.957 --> 00:08:17.270
It really highlights that gap between finding something interesting in the data and being able to consistently profit from it in the real world,

284
00:08:17.301 --> 00:08:17.692
doesn't it?

285
00:08:18.035 --> 00:08:18.551
Absolutely.

286
00:08:19.035 --> 00:08:20.145
Theory versus practice.

287
00:08:20.489 --> 00:08:21.489
The practical hurdles,

288
00:08:21.551 --> 00:08:21.973
costs,

289
00:08:22.067 --> 00:08:22.520
liquidity,

290
00:08:22.598 --> 00:08:23.754
diversification needs.

291
00:08:24.486 --> 00:08:25.588
Shorting difficulties,

292
00:08:25.968 --> 00:08:28.269
they could prevent arbitrage from working perfectly,

293
00:08:28.689 --> 00:08:32.873
allowing anomalies like this to persist longer than you might initially expect.

294
00:08:33.256 --> 00:08:35.154
So the takeaway isn't just about accruals,

295
00:08:35.279 --> 00:08:40.982
but maybe a broader lesson about considering implementation costs and feasibility for any strategy.

296
00:08:41.279 --> 00:08:41.717
Definitely.

297
00:08:42.264 --> 00:08:45.451
It's a great reminder that back-tested returns are one thing,

298
00:08:45.826 --> 00:08:50.076
but net returns after all the real-world frictions are what actually matter.

299
00:08:50.217 --> 00:08:51.857
This deep dive was really insightful.

300
00:08:52.002 --> 00:08:57.829
It shows how market mechanics and investor constraints can explain puzzles like the persistence of the accruals anomaly.

301
00:08:58.028 --> 00:08:58.489
Indeed.

302
00:08:58.970 --> 00:08:59.771
And it makes you think,

303
00:08:59.849 --> 00:09:00.228
doesn't it?

304
00:09:00.634 --> 00:09:06.415
What other known factors or strategies might look good on paper but face similar practical roadblocks that aren't immediately obvious?

305
00:09:06.993 --> 00:09:08.993
Something to keep in mind when you're looking at research.

306
00:09:09.462 --> 00:09:11.743
Thank you for tuning in to Papers with Backtest podcast.

307
00:09:11.744 --> 00:09:13.931
We hope today's episode gave you useful insights.

308
00:09:14.149 --> 00:09:16.212
Join us next time as we break down more research.

309
00:09:16.587 --> 00:09:18.009
And for more papers and backtests,

310
00:09:18.103 --> 00:09:21.149
find us at https.paperswithbacktest.com.

311
00:09:21.623 --> 00:09:22.313
Happy trading!