Appunti completi Economics of Innovation and New Technologies

(D )(L

= − ) + (1 − + )



L S T

t t t−L t−1 t−1

​ ​ ​ ​ ​

−)

Demand without seasonality ( − plus the last forecast without seasonality (−1+−1), that

is the sum of the level and of the trend of the prior period without seasonality

Economics of Innovation and New Technologies 55

Trend

(L )T

= − ) + (1 − 

T L

t t t−1 t−1

​ ​ ​ ​

Seasonality

(D )S

= − ) + (1 − 

S L

t t t t−L

​ ​ ​ ​

α, β, and γ are smoothing parameters for level, trend, and seasonality, respectively.

K* is the length of the seasonal cycle (e.g., 12 for monthly data with yearly seasonality)

F = + + 

L T S

t+1 t t t+1−K

​ ​ ​ ​ +

In general terms we can rewrite it as: = (+ℎ)++ℎ− where h represents the forecast

horizon (e.g., 1 in case you are estimating the next period)

Multiplicative: Use this when the seasonal fluctuations change proportionally with the level (e.g.,

when sales increase, seasonal peaks also increase proportionally)

Level L

(D )(L

= / ) + (1 − + )



L S T

t t t−K t−1 t−1

​ ​ ​ ​ ​

Trend

(L )T

= − ) + (1 − 

T L

t t t−1 t−1

​ ​ ​ ​

Seasonality

(D )S

= /L ) + (1 − 

S t t t t−K

​ ​ ​ ​

α, β, and γ are smoothing parameters for level, trend, and seasonality, respectively.

K* is the length of the seasonal cycle (e.g., 12 for monthly data with yearly seasonality)

F = (L + ) ⋅ 

T S

t+1 t t t+1−K

​ ​ ​ ​ +

In general terms we can rewrite it as: = (+ℎ)*+ℎ− where h represents the forecast

horizon (e.g., 1 in case you are estimating the next period)

Model initialization

Economics of Innovation and New Technologies 56

Accuracy & Distorsion

Accuracy

Forecast Error: the forecast error for period t is defined as the difference between the actual

demand value and the predicted value for that period, so the forecast that you have estimated for

time T.

Distorsion

Tendency of the forecast to consistently overestimate or underestimate actual values.

Average of the errors

If ME is close to zero, the forecast is unbiased on average meaning that there is not a systematic

tendency to either over or under predict.

However, a low ME does not necessarily indicate high accuracy, because large positive and

negative errors can cancel each other out, masking poor forecast accuracy.

Economics of Innovation and New Technologies 57

If ME < 0: Then AD < PD (Actual Demand is less than the Predicted), indicating that the forecast

tends to overestimate the actual demand.

If ME > 0: Then AD > PD (Actual Demand is greater than the Predicted), indicating that the

forecast tends to underestimate the actual demand.

Lecture 7: Forecasting the adoption for radically

new products

You saw forecasting techniques that predict future sales of products that have already been

commercialized and for which prior sales data are commonly available.

We use autoregressive models, assuming that historic (past) sales data are predictive of future

data, with adjustments for trends, seasonality, etc. (i.g. the peak period for smartphones is

Christmas → it is a typical gift)

The same techniques can also be used when launching an incremental innovation, i.e. an

improved/updated product, or a product already produced by a competitor.

We use historic (past) sales data of older products, assuming the new product would sale

similar amounts.

What about forecasting the sales of products that are radically new, i.e. they are first-generation

products, and will create a brand new market?

Brand new product, that doesn’t exist before → no historical data about data and some data are

not useful

E.g., Telepass service in tallroads? No data available.

Product that largely differs from similar product categories. E.g., Tesla (fully-electric) cars in

year. Sales data for traditional cars data unsuitable.

Forecasting in this case is inherently difficult and prone to large mistakes. Even so, we can benefit

from having a systematic approach to tackle these situations. This is the goal of this lecture.

Some attempts to use rogers curve to predict demand, but the main problem with this approach is

that adoption does not depend on demographic features

Examples: the first smartphone, any smartwatch, Netflix, Siri, Alexa and any virtual assistant

Durable → if a product isn’t durable, people switch to another product

Forecasting in this case is inherently difficult and prone to large mistakes. Even so, we can

benefit from having a systematic approach to tackle these situations.

It’s extremely relevant in start-ups → they must do a business plan and they should understand

how to organize

Start crowdfunding as a market test

We can take data about the scale of the market

Empirical regularities

One way to tackle the problem is to learn from past cases.

Economics of Innovation and New Technologies 58

We may encounter empirical regularities, i.e., patterns that occurred multiple times in the past.

Regularities may capture behavior patterns or tendencies, useful in predictions of the future, even if

we may not have a clear causal explanation for what is behind such regularities.

Examples:

Lag between introduction of innovation in racing cars and commercial cars: 4 years

Moore’s Law: number of transistors in integrate circuits doubles every 2 years

10 times law: there must be improvements 10 times greater than the status quo to change a

consolidated customer habit

S-curve

slow pace at inception

pace accelerators at some point: rapid increase

growth slows-downs until it plateu

S-curve

→ is a stable regularity

it is the cumulative sales, not the annual sales → you can’t sell another product to the same person

S-curve of total diffusion (stock) = Bell-shape curve of annual diffusion (FLOW)

Economics of Innovation and New Technologies 59

Why the S-curve pattern?

Having a S curve is sufficient to have a technology that can be useful for some people, but it can be

useless for others

Explanation 1.

Decreasing costs of tech + consumers heterogeneity

Over time the technology becomes less expensive (decreases monotonically). Customers’ personal

value obtained from the adoption is normally distributed (Bell-shaped; central tendency).

As the cost of the technology diminishes, those consumers for which values>cost enter, generating

the S-shaped pattern

Explanation 2.

Invention initially unknown. Consumers adopt when they learn sufficient information about the

technology

Economics of Innovation and New Technologies 60

As learning can occur from consumers who had already adopted, the information about the

innovation increases monotonically. This alone would create the S-shape pattern. This explanation

is robust to the prior one, in the sense that can allow for decreasing cost and heterogeneity of

consumers (it imposes no conditions on costs and consumers).

Tech conditions shaping adoption (Rogers)

Technological conditions affecting the adoption include:

1. Relative advantage of the innovation over existing ones.

2. Compatibility, with the potential adopter’s current way of doing things and with social norms.

3. The complexity of the innovation

4. Trialability, the ease with which the innovation can be tested by a potential adopter.

5. Observability, the ease with which the innovation can be

evaluated after trial.

Social conditions shaping adoptions (Rogers)

1. Whether the decision is made collectively, by individuals, or by a collective body or central

authority.

2. The communication channels used to acquire information about an innovation, whether media

mass or interpersonal

communication.

3. The nature of the social system in which the potential adopters are embedded, in terms of its

norms, and the degree of interconnectedness.

4. The promotion efforts put by change agents’ (advertisers,

development agencies, sales chains, etc.).

Two basic approaches to forecast:

Economics of Innovation and New Technologies 61

Choose the shape of the S-curve that is more relevant to the product categories

Forecast first adoption of a radically-new product

We can reason by analogy with innovative products of the past

The model should fit an S-curve (stable an empirical regularity)

Bass Model: It is model that explains adoption by focusing on how potential consumers decided to

adopt.

Bass posits that there are 2 basic forces behind adoption:

1. DIRECT ACTION OF PRODUCER through media or vendors’ communication (e.g. press, shops,

advertising, vendors chain)

2. WORD OF MOUTH through acquaintances and social contacts (e.g. friends, neighbors,

strangers in street, etc.)

IMPORTANT: Bass models only the first purchase of a new product

Bass model

The Bass's model is an epidemic model, that is, it draws by analogy on patterns of epidemic spread.

BASS MODEL:

Animal-human → DIRECT ACTION OF PRODUCER → captured by parameter

p

Human-human → WORD OF MOUTH → captured by parameter q

Probability to adopt

Economics of Innovation and New Technologies 62

Annual adopters (= annual sales)

The sales of one year (t) can be obtained by multiplying the probability to adopt, times the

remaining market, i.e. the potential market has not yet adopted.

in t=0: Y_t = (p+0)*N

The merit of the Bass model is that it is elegant (parsimonious) and uses 3 parameters that have an

easily-understandable meaning.

m >> total potential market

p >> Media and vendor’s communication parameter

q >> Imitation parameter (word of mouth rate; infection rate)

Implications of the model for marketing strategy

If you having a strong marketing campaign → you will have an higher p

At the beginning, the media and vendors’ force is always critical, because total sales Y(t) depend

only on p at t0.

Marketing strategies (e.g. strong vendors push, advertising, ..) at launch is key to lift every

product launch.

When about half of total market has adopted (Nt-1 ≈ m/2), vendors’ communication is less

critical.

Social contagion (word of mouth) is effective only after reaching a sizable customers’ base.

Products that are more visible, can be the subject of friendly conversation, or involve network

effects are more easily diffused by word-of-mouth

When the media and vendors’ communication component is modest (p is low), projecting high