Going through another model today, and I was projecting out subscribers and rev. per subscriber. This time in particular, felt like I was basically shooting in the dark with a blindfold on, +/- whatever management and the street are saying. AKA zero conviction
Lets say you don't have any fancy data resources; no alt. data, no paying for any data-providers, and somewhat limited access to sell-side datasets.
What is your process when it comes to bottoms up forecasting the KPIs that drive revenue for a company? In the absence of really strong data assets, how do you try and build a more structured process around modeling the drivers for a company?
I think one issue I have struggled with, is that if I rely on historical numbers/trends + common consensus (this industry grows x%), I am likely to wind up in the common consensus bear/base/bull ranges.
I think a good example would be a case study at a hedge fund. How do you not end up in the same place as consensus or "fair value" with only the basics in 4hrs (10k/qs, sell side reports, earnings calls, etc.). Perhaps the best case outcome is there is something nuanced in the growth algo / profit algo of the model that you can catch a reinforcing or cliff moment that is likely to swing perception one way?
Trending Content - Hedge Fund Forum
| +49 | Do Quantamental roles really exist? | 24 | 6d |
| +17 | L/S Equity Career London | 11 | 3d |
| +13 | Hedge Fund Pivot | 3 | 17h |
| +11 | Buy Side Credit - Seats / Guide | 1 | 1d |
| +11 | Mid-career shift to buy side with niche background | 3 | 1d |
| +9 | Nvm | 6 | 3d |
| +9 | Wfh / work remotely at MMs or SMs? | 5 | 20h |
| 0 | Hedge funds | 1 | 23h |
Career Resources
Based on the most helpful WSO content, here are some strategies to improve your forecasting and build a more structured process around modeling the drivers for a company, especially when you have limited data resources:
Understand Revenue Drivers:
Historical Trends and Key Drivers:
Competitive Positioning:
Bottom-Up Analysis:
Cost Structure and Margins:
Balance Sheet and Cash Flow:
Alternative Data Sources:
Nuanced Analysis:
By following these steps, you can build a more structured and informed process for forecasting, even with limited data resources. This approach will help you move beyond common consensus and develop a more nuanced understanding of the company's revenue drivers and growth potential.
Sources: On the Job With Simple As… My Research Process, Give me your best examples of first hand research / edgy view!, Associate guide on analyzing an income statement, How to analyse business models?, Q&A: HF Analyst @ $5bn+ Fund - Breaking In and Transition to Risk-Taking Role
Subscribers and rev. per subscriber - depends on the company. Things like that are usually linked to consumer income on the disposable side unless it's a necessity - historical numbers are BS. It's all down to what industry you're modelling - stop being a quant focus on fundamentals. Fun fact - no one in the sell side knows how to model the NVIDIA growth rate because the growth is stupid high and this type of expansion hasn't happened before so everyone is thumb-sucking.
Me personally - when I was forecasting it would be the case of asset yielding acquisitions - asset yielding disposals and getting granular - I caveat I worked in hard assets so a bit different - that's my generic answer to your question unless you have a good example of what you're trying to model.
Underrated comment “stop being a quant focused on fundamentals.” This is an error I’ve seen a lot too. The world is outside your models. No amount of data can tell you how many people are going to subscribe to Netflix. The #1 way I got better at forecasting is reading books on decision making. This includes things like philosophy, probability, psychology, etc.
Would you be willing to share any particular books?
(1) use historicals as far back as possible. Especially in cyclicals. Annual is ok to start. Too many ppl look 3/5/10 years back. If you haven’t seen how the nuts & bolts interact during a downturn, how volume / price / etc interact, you will fail to anticipate the non-linearities that are key.
(2) try to understand, model & think about the business from an operator standpoint:
- FIRST: What is NOT under mgmt control (macro, competition, industry s/d)?
1a) What kind of visibility does mgmt have into how these items trend.
1b) What’s a realistic range of potential outcomes?
ex: Is the company 3% of supply & capacity takes 5yrs to bring online? Quite visible. Or are you a disaster relief company with politicalized contracting process and unpredictable annual TAM?
**** if you can’t answer these, stop. There is no short cut. You don’t have to know better than mgmt, but you do have to understand what mgmt knows.
SECOND: given this range of external potentialities:
- 2a) what are mgmt’s incentives & targets? (5yr growth? Mkt share? Community adjusted ebitda? Topline growth? Stock price?)
- 2b) what decisions are available to mgmt? Of the main decisions (expand/cut capacity, utilization, staffing levels, pricing decisions, new verticals, m&a, capital returns)
-2c) for each potential decisions, bridge the (2b) actions to the (2a) incentives. How much could each action impact metrics? How long does it take? What is the downside risk or degree of certainty it works?
**** here you’ll likely find most of your diligence. Does GM discontinue electric car division? How much would that help (h what’s the margin difference vs other segments, or capex savings?) how long would that take (can the plant/staff be used for ICE models or will there be costly restructuring? How much capex investment is sunk vs saved? Can it be restarted if there are new govt incentives? What are the union barriers to shutting that down?) or perhaps (can I raise price? Basically - here you have to do the typical customer/competitor surveys/calls/tracking etc that are generally discussed)
this is kind of rambly but I’m gonna stop here cause these imo are pretty much where ppl ‘don’t get it’.
The sell side is incentivized to pretty much reflect mgmt guidance or basically to back into a PT a bit higher than current in out years. they are not doing this work in most cases.
Depending on SM vs MM or generalist vs sector, your path to answering these will rely in Differing weights on:
- brute force grunt work like typing in 100 months of monthly auto sales from PDFs, - throwing institutional resources at shorter-term data
- savvy fieldwork and cultivation of primary sources, like recurring calls with private competitors / suppliers or 135 representative car salesmen
- really creative stuff (hard to explain, but it feels like a short cut & only lasts a while)
- my favorite: inverting, eventually once you know a few dif business models and get some reps (or read enough), you’ll begin to see the common signs of everyone else misforecasting. Imo it’s a lot easier to prove why the prevailing forecast is likely wrong, than to make a better one. Proving the alternative view wrong is equally sufficient
You should write a book
I’m glad you found it helpful. I’d love to write a book someday, but I’m not full enough on useful knowledge to fill one. Anytime I think I’ve conceptualized something new and potentially useful, I quickly find it’s been said many times before in clearer & more complete way! But that’s the fun. Maybe someday.
I missed that you’re asking in a 4hr case context, so I’ll adapt the above:
Just give a little moment of thought to each to each one of those items (prob comes up in chat) and then hopefully you’ll see 1-2 external uncertainties & 1-2 company changes/decisions that jump out as clearly being the main unknowables w/ a large range of uncertainty & a large impact on earnings.
Go ahead and just make the full model & tune it in line with Cons/Guidance for the Base case (tweak base case at the end if time allows)
Now for ur Valuation, just use multiples & make it as easy for urself as possible (hopefully FY+1 or FY+2 EBITDA basis if that’s appropriate for the story & sell side / comps typical range)
Let’s say example: base case (full model off guidance spits) sales growth from 907 to 1000 for FY+2 (5% cagr) & margins expand 100bps over 2yr 10% to 11% so BASE case FY2 EBITDA = 110 from current 91m.
Use fcf/capex guide or est to bridge current BS to FY+1. Pick ur mult (don’t diverge from today’s mult in base case w/o a very good reason)
now for bull & bear case, u really just need to show that u understand the important levers & how they generally interplay & a reasonable range. So make up a good scenario w/ rev*mrg impact for ur bull & bear case to get a defensible bear/bull mult that u can tweak if ur debrief reveals the assumption was stupid. Ex:
*2pt annual rev tailwind vs base case thru 2025* & or ~40m incr rev*
[new product launch is modestly more successful than expected]
OR
[important regulatory vote passes & accelerates share]
*addl rev comes thru at 50% incremental ebitda mrg*
[cause this product is 20% higher GM vs base case product mix + overhead absorption]; or [new rule accelerates upgrade cycle & mkt can’t add production for years, so pricing is stronger & drops to ebitda]
So bull case EBITDA is 130m vs 110m or 40m incremental rev @ 50% incr margin
*do u need a different multiple for this scenario? Think about if the driver changes are durable beyond FY2, if they represent a cyclical peak / trough, or a biz mix more valued by mkt as higher quality*
- u have Bull case EV
* how does BS look dif in FY+1 in this scenario* think about interim net cash accumulated from this new ebitda. Don’t do too much here unless addl capex, WC intensity, etc is likely material*
*repeat for bear case*
ok now you have base case model that says [I can model], and u have base case tuned to mgmt guidance / consensus.
You’ve shown u know the *key debate / question* that could drive upside & a generally not-dumb understanding of what that means for earnings/multiple & stock price. Same for the key risks shows in the bear case.
Slap a probability on those cases 30% 45% 25% ***elegant way to say… reminder I have no convicted variant view*** if interviewer comes in super beared up, just make say ahhhhh I can see how bear situation might be more likely; if I change these to 15% 45% 40% my blended PT moves down 20% and it’s a great short!!!!
Now your only real risks are: did u totally f’ing miss the biggest upside/downside drivers? That’s really the one that matters.
Other snafu’s are: Did y get them right but pick unrealistic assumptions? Did u pick reasonable scenarios but misunderstand the eps flowthru & multiple implications? Did u pick prob distribution at odds with ur interviewers strongly held views? You can whiff on all of these & probably even cut big corners on the model and be fine
As an entry level interviewee - just showing you ‘get it’ & you can frame up a problem quick & identify key questions to diligence shld put you ahead of most. The rest is learnable. If you’ve covered this sector on sell side or prior HF role, obv less leeway for dumb mults & assumptions
so basically - all we did here was your existing consensus outlook base case you already do, slapped on some scratch math de-commit to our ‘thesis’ and refocus toward showing we can ID the key Qs & next steps.
One of the most helpful comments I have read on this site. That said...how do you get all this done in 4 hours??? I just had a case study on Monday where it was the full model build out + a write-up--investment summary, variant view and difference vs estimates, key drivers, risks and mitigations--all in 3 hours! And I drilled modeling enough where the 3 FS were done in 2 hrs, but modeling the scenarios in my revenue build essentially took the remainder of the time (and yeah I did a DCF bc it was an LO but I should have just done multiple valuation)....I guess my takeaways are to get 3 FS done ASAP in 2 hr, do multiple valuation, and then just BS the scenarios haha. I honestly can't help but feel a bit pissed off, because I know this stuff and can deliver quality work if given the time, but 3 hours is just outrageous.
Edit: I went to the next round somehow
How do you allocate cash on the BS? You can’t predict M&A/debt repayment/SBB.
Do you just grow cash position?
Also, even for companies I’m bearish on, I tend to have a growing ROIC because I don’t know how project invested capital. It’s an issue.
Thanks
Extremely useful stuff
Would you mind elaborating in a future post or comment about the "really creative" stuff or give an example? Sounds fascinating, and I hope I can practice it with my limited college student time haha
Culper cough
To your last paragraph, who said that the right answer in doing an interview case study is not the same as consensus?
You guys make it too hard on yourselves. They are testing to see whether you can produce standard work and distill the basics under pressure and under limited time. Of course no one expects a differentiated channel-check or such in that amount of time.
I skimmed over OP and didn’t realize it was a case study. I thought OP was asking in reference to Y1 struggs.
Really? I agree that - a solid grasp of the key questions / next steps & standard demonstration of table-stakes modeling is probably sufficient. But personally I’ve only gotten offers historically in cases where I expressed a variant view. To be fair, when I couldn’t find one I def spun wheels and did a junk job on the basics as well.
Maybe it’s a dif SM vs MM, but I kinda feel like an SM wants to pick a name that they know well enough to drill down in Q&A… and if they aren’t long or short or looking, they won’t know that name. So they probably think consensus is wrong just by virtue of it being a name they know well enough to Q&A on.
To MMPM’s point though OP, no one’s holding you to your forecast. You can just take a random ass slightly contrarian view in the main debate vs consensus, then leave yourself a (totally fair) cop-out ‘but I’d diligence XYZ to make sure my random-ass contrarian call on the key driver is right. If further diligence doesn’t prove that, then it’s probably closer to consensus earnings in which case it’s [fairly/unfairly] valued.
Really great feedback I appreciate it! Also I think this summed up doing the case quite well:
"so basically - all we did here was your existing consensus outlook base case you already do, slapped on some scratch math de-commit to our ‘thesis’ and refocus toward showing we can ID the key Qs & next steps"
I'm not at a fund and not on a realistic path to working at one, but I do public equity investing so always enjoying learning + off chance a small firm is interested in me at some point.
Obv recognized that sell side modeling just ends up as a reverse engineered management target/guide, and in the specific stock I was working on, I technically had more than a shot in the dark (sub growth is a by-product of aggressive marketing spend, new markets/products, + would be best to do cohort analysis probably). Key issue for stock is 1 big thing and I might've been getting cute by trying to justify a thesis to disregard market's focus on semi-binary event and underwrite everything else... usually doesn't end well.
Putting that aside, the general spirit of the question has come to me at times, and I wanted an opportunity to hear from those on the job on how they tackle forecasting thoughtfully. + general thoughts on a case scenario since that is helpful itself, and it has some similarities to my on the job stuff in terms of restricted resources + time as we don't dedicate a ton to proprietary research.
Anyways really good stuff and I appreciate the detailed response
I mean, either way, you are not doing anything more than some back of the envelope estimating. Either way, you are taking massive stabs in the dark.
This is merely a consequence of the standard top-down approach to modeling. The standard financial modeling is to use a computer as merely a calculator that spits out outputs from inputs you impose on it. You aren't using a simulation where the inputs arise organically from the model itself.
More important than what any given number is, is understanding what the key inputs are and how variations in those inputs flow through the model. Getting that "fair market value" is relatively straightforward and, for the most part, can mostly be done with "only the basics in 4hrs", to borrow what you said. The key thing is understanding how much things can change from there. That's where the value add largely comes from in such top-down modeling. You will all get relatively the same number.
The question is do you know what the drivers are and how much they can actually drive? And are you comfortable with the outcomes of those ranges
Yes good points I agree - big part of modeling is just to be able to flex your business model and inputs across scenarios to better understand what is being reflected at current price, and how valuation + EPS/EBIT/FCF/etc. changes as these things flex. In that vain, the "base" case with the "basics" and guidance + etc. usually gets you around fair value anyways - obviously.
I think mostly what I was getting at is increasing my conviction on the varying outcomes / KPIs / drivers - which probably just ties back to really basic stuff like: how well do you understand the business and industry and what management is attempting achieve. Also there is an element of knowing what you can and can't diligence, and knowing where to place your bets I suppose.
Agreed.
A large part of what I said is really just how I wanted to vent about how much I despise the way most financial modeling gets done because you are coming in from the top and imposing on the math the inputs that in reality you want to have the model solving for. And, since that isn't how most financial models work, it really just means who cares about the convictions in the numbers. It is much more important to understand how things can be different and how those differences can flow through the financials. Which, as you point out, is really "just" understanding the business/industry, the management team's objectives, what can't be known (although a real bottoms-up financial model makes this far less relevant), etc.
Getting better at forecasting requires thinking divergently and leveraging inside and outside views to calibrate your forecast. Here's a run through of how I would approach forecasting revenue for a company but I think what you're really looking for is scenario analysis:
1) Know the business and industry -- understand the growth algo for the industry, e.g. if looking at EV what's the adoption rate (penetration rate) of EV vs ICE over an X period of years. I might tie it to regulatory factor, like COP targets/goals, because big government policies tends to be large structural headwinds/tailwinds for industries. If most of the developed world has a goal of 50% EV-to-ICE by 2040, and current ICE production is 75mn units per year growing at 3%, it's easy to get an estimate of unit volumes through your projection period (75mn * 1+3% ^ 16 years). Then, apply the 50% penetration rate to get to a target EV unit production number. Stress the assessment by comparing it other studies/perspectives. Forecasting is not about predicting the future. We are trying to determine what is the most likely outcome based upon current information. Gotta be Bayesian.
2) With a quick and dirty TAM done, look at current EV vendors -- TSLA, RIVN, BYD, along with legacy OEMs -- and lay out the competition positioning of each firm. This really boils down to three distinct things: (a) do they have a product and is it good or bad, (b) what is the current industry dynamics e.g. is there a price war or a supply issue, and (c) what is management's plan e.g. if they are pureplays they will need to get to 10-20mn production levels within a 5-20 year period (level of legacy OEM production), do the companies lay out their plan to get there, and if so, what is it?
3) Build the management case by integrating the specific items management is targeting (e.g. "we have built 2 car factories producing 500k cars each and are on track to build another facility by 2027"). You can "average" those out through your forecast period -- e.g. new factory every 2-3 years. At this rate, the company will have X capacity by 2040. Note, at this point the only thing in your model is the TAM outlook (CAGR, and math that includes today's production #s and 2040s, along with the factory/capacity targets).
4) Now we can build out an annual model to forecast fiscal year sales. We'll do this by first applying the CAGR. This is naive and is specifically incorrect, but it's how you'd start. Next, look at historical product cycle and seasonality of legacy firms and synthesize it with the target company's plan -- e.g. new product refresh every 2 years, new car type every 3 years, currently only has 5/9 categories filled. Capacity has important consequences here, acting as a limit and leaving a noticeable impact on your estimates. Use an average of unit seasonality (due to factory retooling, consumer demand, etc.) with product cycles to figure out your base case for y/y growth. Now, modify the basic CAGR with updated views on y/y growth incorporating seasonality and product cycles.
5) Next, calculate the quarters of the next 2 fiscal years (minimum of 8 quarters so you can do NTM+1). For the quarter, do the same seasonality and product cycle analysis but on a quarterly basis. E.g. production is heaviest in q3-q4 and sales greatest in q2. Use what you know about the company to incorporate information about specific events, such as a product launch (e.g. teasing of a new SUV by end of FY 2025).
Up to this point, you're tracking consensus. To go beyond consensus, we need to explore scenarios and find the edge cases.
6) We're now ready to build scenarios and add in granular details. Scenarios should first be exploratory, such as macro scenarios ("rates lower next year" or "recession outcome") and company specific items (product success/failure). Granular details can also include factory updates, new build outs, regulatory credits, and other potential impacts. You can also incorporate competitive dynamics, e.g. "what would a price war look like?".
7) Once you have explored a few scenarios, create a probability distribution (e.g. "strong buy case is where co becomes a market share winner -- I think the probability of that outcome is 65%"). Your model should reflect the scenario you think is most likely, and from then on, your "coverage process" revolves around updating and calibrating that view with incremental information (Bayesian). Scenarios need to be attached with specific "print dates" (when will you know it's true/false) and catalysts (when/why will the market/street come to your view?). E.g. "company's new product launch in q2 nfy will boost their market share to 80% as they are on track to reach 8.5mn in run rate by end of fy25" tells me that the KPI you are betting on is their annual production, and the catalyst path includes a near-term product launch, which should close the gap between your expectation and the streets.
Note: for your scenarios, you should have a clear understanding of the qualitative drivers and quantitative impacts (to the KPIs). For example, if you think a new CEO will come in and pull forward the launch of a new car = X% increase in units x ASP = delta to sales = new sales estimate (or, sales estimate of a scenario).
Helpful
There is no fucking way any 3p is putting together better forecasts than the management team.
They are ALL shots in the dark. Yeah there are some things you can do to improve but IMO it's generally a huge waste of time.
What is 3p?
I may not have more data than management, but I’m trying to be accurate & they’re trying to make share price go up.
I also might be less biased, as I don’t anchor my life to the forecast outcome.
I have the advantage of meeting all if managements peer companies from I may get a view likely competitor plans whereas management obviously doesn’t do 1x1s with each other
And beyond that - I’m guessing you don’t work with many smid cap industrials but it’s entirely possible my forecasts can be more accurate, sometimes they’re more operators than strategists / forecasters
When you model subs you should usually do it as a net change number in actually subs. So qoq or yoy change in net adds
Agreed, but OP needs to first get down to a fundamental view on what's happening. E.g. "NFLX user strategy was successful in LATAM, if it can achieve that penetration rate, it can increase its sub count from A to B by X year" to "current estimates on X year are 25% below my scenario" to "NFLX will begin to report new subs through this program in Q2 of next year, between now and then I'll take consensus as channel checks don't point to any variance this qtr or nfq" to "starting in q2 NFY, I model in XX customers which is Y above the street".
It sounds super granular at the end but it is driven by a view of a scenario you have assessed as likely at some point in the future.
Ahh, you think modeling is supposed to predict the future. It's fugazzi. You should have a target revenue and EBITDA in mind. Then just tune with anything reasonable that hits those numbers.
You need to model out to consensus and use guidance and CC commentary around guidance or use the sell side models. You're trying to figure out what is baked into current numbers and then make a call on the achievability of those numbers given the assumptions baked in. If numbers look wildly unattainable then you have an actionable situation. and vice versa
what is cc commentary? And why model to sell-side numbers when buy-side expectations is what matters (even though its every hard to gauge for any given quarter and even harder beyond that)?
CC is probably channel checks. "buyside bogey" probably just as important, but thats the second level after getting the SS consensus figured out. Just simplifying the process here
Diligencing the achievability of consensus / buyside / stock price implied results is the goal.
Minus nam atque molestiae quos ullam provident. Vitae consequatur nulla aliquid recusandae. Voluptate maxime rem sit distinctio quod.
Et et quidem eveniet. Tenetur debitis sed itaque dolores est accusamus dicta. Velit ea vitae nostrum nam inventore facere. Eligendi omnis ea eligendi rerum et culpa ipsum facilis. Est et quia non repudiandae consequatur laudantium.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...
Non et quam vero quod. Nihil voluptatem in est sequi. Assumenda quia delectus molestiae quibusdam ut voluptatem saepe. Perspiciatis qui repellendus et. Qui accusantium similique quia nesciunt. Aut qui provident enim mollitia.
Adipisci officia quis repellat nulla ratione. Nisi similique nemo rerum qui dolorem. Et reprehenderit odit sint aut illo.