Just parsing through
Previously I had been using the programming language “R” to parse through company announcements and comparing them for clues on future performance. I’ve been using decision trees and random forests to see if companies that have performed well over the last 5 years use similar language.
Linguistic comparisons between companies are interesting, but the same company can often shift from being a great investment to well below par. An example might be PZ Cussons, which listed on the London Stock Exchange in the 1950s at a value of just over £1m versus around £850m today (a 850x bagger… and that is not even including dividends!) Yet over the last 10 years the PZ Cussons share price has trod water, and the value now is the same as 10 years ago.
The PZ Cussons reported numbers have obviously disappointed relative to expectations, but surely the financial commentary ought to manage expectations that tougher times are ahead. So rather than comparing the language that difference companies use, perhaps there is an opportunity to use Natural Language Processing to compare how one company’s announcements change over time. My approach is to treat announcements as stepping stones – share price positive or negative jumps from one statement to the next and on to the next. Investing is an expectations game and this approach reflects that.
Top heavy with intangible assets
Rather than PZ Cussons, I’ve focussed on Hostelworld because the company’s balance sheet is so top heavy with intangible assets. Intangible assets are €112m, 5x greater than tangible assets of €21m and a stockmarket value of £127m. Moreover those intangible assets recorded at cost are almost €300m according to note 11 of the Annual Report, consisting of €47m of goodwill paid for other competitor businesses, €214m for domain names and €30m of other intangible assets, including affiliate costs and software. €300m of intangibles are not recorded on the face of the balance sheet because over time Hostelworld have amortised and written off €163m of intangible assets…that is more than the current market cap of the whole business!
HSW is a platform business – connecting hostel travellers pass through with somewhere to stay for the night. The website / app relies on user generated content (ie reviews by fellow travellers) – I rather like businesses where the customers contribute to the success of the business (and pay in advance!) Hostelworld just provides the booking platform where customers reserve their bed. So an Online Travel Agent like booking.com or expedia.com but on a much smaller scale, because it focuses on hostel travellers, a smaller market than hotel guests.
The website has been going since 1999, but it only listed with an IPO on the London Stock Exchange in October 2015 at c. 190p. The share price peaked at 422p in March 2018 and then fell to a recent low of 107p in October 2019. That’s a 75% decline from the peak to trough.
Network effects and unstable demand
Management have limited visibility about how many customers are going to go travelling in the peak summer season and use their site / app. This explains how the share price can be so volatile. In the following commentary I would like to make clear nothing untoward is going on – just because the share price falls 75% that reflects the inherent uncertainty of a platform business with network effects and unstable demand. I’m not picking a fight with the company, I’ve actually bought the shares close to the recent lows. Instead I think that Hostelworld management updates are fertile area to practice Natural Language Processing on.
Kylie Jenner’s intangible assets
The financial numbers do also tell a story, with the business trading on a very high price/tangible book but a very low price/free cashflow. Like many businesses, there is not much connection between capital invested in property, plant and equipment (tangible book value) and market value (determined by expectations of future free cashflow). Just ask Kylie Jenner who has sold a majority stake in her cosmetic’s business for $600m. So rather than the history of reported financial numbers, I’m interested in how the company manages future expectations with announcements.
Click on the image to enlarge.
The table above shows the major announcements made by Hostelworld in chronological order (oldest announcement at the top) since it came to the stockmarket with an IPO in October 2015. The data comes from the quantmod package, and also webscraping using beautiful soup in python (Thanks Tom!). On the right you can see the share price open on the day of the announcement, and the far right column is the share price movement after the announcement until the next announcement. I’ve colour coded the movements with green very positive and red very negative. Shades of yellow / orange are somewhere between.
You can see a gradual progression higher from the IPO until March 2018, when the share price peaked at over 400p, and then a decline to recent low 18 months later of 111p in October 2019. In the middle of May 2018, very close to the peak valuation, the company announced that the Chief Executive was leaving, ironically at the time claiming “He leaves behind a business with significant competitive advantages and momentum.” We shall see that subsequent guidance and share price performance means that this claim should perhaps not be taken at face value. Click on the image to enlarge.
The second table is exactly the same data, but sorting the announcements from most negative (red) to most positive (green) in terms of subsequent share price performance. Very often the most informative announcements seem to be very laconic AGM statements made in May or early June, managing down expectations with perhaps only one or two sentences. This in itself is an informative signal, often the most price sensitive information (disappointing expectations) is made in as few words as possible, with little adjacent explanation.
Quanteda Key Words In Context
I then used Quanteda* to do some basic Natural Language Processing analysis. One of the interesting features from Quanteda is that it is possible to search a corpus of company announcements using the KWIK function (Key Words In Context). Anyone who has studied Theology will see this is a concordance tool. I used the package to search on “competition” – see the appendix below for the full results.
Signalling competition
In 2015 and 2016 no announcements mentioned competition. In 2017 there was just two mentions of the word (one each in the Interims and Full Year results). Then in 2018 and 2019 discussion of competition greatly increases as Online Travel Agents like booking.com attempted to move in on Hostelworld’s space.
The Full Year 2018 announcement (share price 368p) the company mentions “competition”, “compete”, “competitive”, “competing” 7x, without actually warning that competition could feed through to reduce expectations. Management wait a couple of months until June 2018 to say:
“While competition has increased into Europe and we face headwinds from adverse foreign exchange movements and we have limited visibility on bookings in the forthcoming peak season we reaffirm our expected earnings estimate for fiscal 2018.”
So parsing this: in June, the company begins to warn about competition, but then rather contradicts this warning by saying numbers for 2018 shouldn’t be affected. The contradiction is resolved by the share price reaction -21% from June 2018 to the H1 announcement in August, from when the shares reacted by falling another 31% (from 270p to 185p) over the next few months. By this stage it should be obvious that competition IS having an impact on financial expectations. Then a year later in April 2019, with the share price below 200p, the company admit:
“In recent years the use of third party platforms has increased with these platforms typically providing connections to multiple OTAs, including Hostelworld. This has allowed more competitors to gain access to the hostel market. Therefore upgrading our third party platform connectivity is a key priority in order to defend our competitive position, ensuring we have the best access to inventory for our customers.”
Very clear – companies often feel that they can be more candid after expectations have been reduced. I’m not sure why this is – I think that companies would gain more credibility with investors if they were more open about potential threats to their business, before things start going wrong.
Conclusion: read between the lines
In conclusion, it seems that there are clues to future performance by analysing language. But neither human readers nor computers should take management communication at face value, and instead “read between the lines” and look at what is implied but not spelled out boldly in print. I feel like there could be some useful Natural Language Processing techniques to help do this.
I’m still playing with the Quanteda package, and hopefully should be able to do some N-grams analysis and maybe also combine it with SpacyR to do Named Entity Recognition. I’ve even bought a book on Keras (a deep learning framework) but I’m not clever enough to get very far with it. I had previous published this bearish post on hostelworld in 2016. But now I’m long, just because I feel that the bad news (threat from competitors) has been communicated and is now reflected in the numbers. To be clear, none of this should be taken as advice. Do Your Own Research. But do get in touch with me on twitter @brucepackard
* https://quanteda.io Benoit K, Watanabe K, Wang H, Nulty P, Obeng A, Müller S, Matsuo A (2018). “quanteda: An R package for the quantitative analysis of textual data.” Journal of Open Source Software, 3(30), 774. doi: 10.21105/joss.00774,
Appendix Key Word In Context “Competition”
Date | docname | from | to | pre | keyword | post | pattern |
---|---|---|---|---|---|---|---|
28 Oct 2015 |
Publication of Prospectus |
||||||
21 Jan 2016 |
Pre-close Trading Update |
||||||
05 Apr 2016 |
Preliminary Results 2015 |
||||||
26 May 2016 |
AGM Statement |
||||||
23 Aug 2016 |
Half Year Report |
||||||
25 Jan 2017 |
Pre-close Trading Update |
||||||
28 Mar 2017 |
Final Results 2016 |
properly focused on the customer and maintain our |
competitive |
position in our core markets . We will |
compet* |
||
01 Jun 2017 |
AGM Statement |
||||||
22 Aug 2017 |
Half Year Report |
1 ) to invest in our Core Products |
competitiveness |
to ensure our platforms are the preferred choice |
compet* |
||
17 Oct 2017 |
Capital Markets Day |
||||||
24 Jan 2018 |
Pre-close Trading Update |
||||||
10 Apr 2018 |
Preliminary Results 2017 |
substantially improve our offering to customers and our |
competitive |
position and underpin the Board’s confidence that we |
compet* |
||
10 Apr 2018 |
Preliminary Results 2017 |
the inherent growth of the sector and to |
compete |
more effectively with generalist online travel agents ( |
compet* |
||
10 Apr 2018 |
Preliminary Results 2017 |
competitive |
position . Â The pilot launch resulted in |
compet* |
|||
10 Apr 2018 |
Preliminary Results 2017 |
the objective of differentiating Hostelworld’s app from the |
competition |
whilst also driving app downloads , Speak The |
compet* |
||
10 Apr 2018 |
Preliminary Results 2017 |
properly focussed on the customer and maintain our |
competitive |
position in our core markets . We will |
compet* |
||
10 Apr 2018 |
Preliminary Results 2017 |
objectives will be : 1 . Â Â |
Competing |
on Core Product Functionality and Differentiating USPs . |
compet* |
||
14 May 2018 |
CEO Succession |
competitive |
advantages and momentum . We wish him every |
compet* |
|||
11 Jun 2018 |
AGM Statement |
competition |
has increased into Europe and we face headwinds |
compet* |
|||
21 Aug 2018 |
2018 Interim Results Announcement |
market , particularly in Europe , is increasingly |
competitive |
. In addition the World Cup and the |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
in July this year will further enhance our |
competitiveness |
. We are increasing our technology investment in |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
Porto to further improve and differentiate us from |
competitors |
. Hostelworld has the ability to continue to |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
1 ) To invest in our Core Products |
competitiveness |
and Differentiating USPs to ensure our platforms are |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
move for the business which will enhance our |
competitiveness |
. As with our other booking options , |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
whole , external geopolitical factors and an increasingly |
competitive |
marketplace have changed the Group’s risk profile in |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
competition |
, and the proposed exit of the United |
compet* |
|||
21 Aug 2018 |
2018 Interim Results Announcement |
. Â The business operates in an increasingly |
competitive |
marketplace and our relative scale and size could |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
and technology change . We have seen increased |
competition |
from other online travel agents ( ” OTAs |
compet* |
||
21 Aug 2018 |
2018 Interim Results Announcement |
feedback from customers as to the relevancy and |
competitiveness |
of our proposition as well as propensity to |
compet* |
||
29 Nov 2018 |
Strategy update and trading update |
; – Â The strategy is to enhance |
competitiveness |
and deliver a return to growth by focusing |
compet* |
||
29 Nov 2018 |
Strategy update and trading update |
strong foundations of the business , increase our |
competitiveness |
and capitalise on the significant opportunity to take |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
programme . Hostelworld is operating in a highly |
competitive |
market , which is growing . We have |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
and native app development strengths and reaffirm our |
competitive |
position as a leading hostel focussed online booking |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
as a market leading category specialist we can |
compete |
effectively with the generalist lodging online travel agents |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
to be addressed to ensure that we remain |
competitive |
in our marketplace . These will be the |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
experience of our customers bringing it up to |
competitive |
parity . Since 2015 there was underinvestment in |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
, which was insufficient to maintain our platform |
competitiveness |
against other OTAs . I have identified closing |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
, including Hostelworld . This has allowed more |
competitors |
to gain access to the hostel market . |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
a key priority in order to defend our |
competitive |
position , ensuring we have the best access |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
and recognition . However , in an increasingly |
competitive |
market where our customers have more choice and |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
Hostelworld has developed are a key source of |
competitive |
advantage for our business . ·   |
compet* |
||
02 Apr 2019 |
Preliminary Results 2018 |
. Outlook Hostelworld is operating in a highly |
competitive |
market , which is growing . We have |
compet* |
||
31 May 2019 |
AGM Statement |
. Hostelworld continues to operate in a highly |
competitive |
market where weak consumer sentiment in key EU |
compet* |
||
21 Aug 2019 |
Interim Results |
hostel partners . Â The market remains highly |
competitive |
and as noted in our AGM statement , |
compet* |
||
21 Aug 2019 |
Interim Results |
experience for all our customers and close the |
competitive |
and technological deficits which we had previously identified. |
compet* |
||
21 Aug 2019 |
Interim Results |
year , which will further strengthen our inventory |
competitiveness |
.  ·     Continued |
compet* |
||
21 Aug 2019 |
Interim Results |
Growth ‘ , the business will be significantly |
competitively |
strengthened and well-positioned to capitalise on the growth |
compet* |
||
21 Aug 2019 |
Interim Results |
.  Outlook  The market remains highly |
competitive |
and as noted in our AGM statement , |
compet* |
||
21 Aug 2019 |
Interim Results |
Hostelworld brand continues to operate in a highly |
competitive |
market , which has also been combined with |
compet* |
||
21 Aug 2019 |
Interim Results |
experience of our customers bringing it up to |
competitive |
parity . As a result , the Group’s |
compet* |
||
21 Aug 2019 |
Interim Results |
whole , external geopolitical factors and an increasingly |
competitive |
marketplace have changed the Group’s risk profile in |
compet* |
||
21 Aug 2019 |
Interim Results |
the most significant of these factors are increased |
competition |
, and the proposed exit of the United |
compet* |
||
21 Aug 2019 |
Interim Results |
. Â The business operates in an increasingly |
competitive |
marketplace and our relative scale and size could |
compet* |
||
21 Aug 2019 |
Interim Results |
and technology change . We have seen increased |
competition |
from other online travel agents ( ” OTAs |
compet* |