FAQ: BRIGHT MARKET INSIGHT
What is Bright Market Insight?
Bright Market Insight are unique proprietary developed forecast models for Pulp (prices for NBSK, BHKP, Fluff, Dissolving, Viscose, Linten Cotton in NA, Europe and Asia) and Timber (red- and whitewood log prices in Scandinavia) based on artificial neural networks (ANN).
Who is behind Bright Market Insight?
Bright Market Insight is owned, developed and produced by Calejo Future Intelligence AB. The company was established in 2013 to guide the processing industry to better strategic decisions about the future and better strategic investment decisions with the support of proprietary produced forecast models designed on the basis of artificial neural networks.
Early Calejo was named a "Vinn Nu" (win now) company by Vinnova and the Swedish Energy Agency (Energimyndigheten) in competition with other start-up research and development-based companies. The company also won a prize in Venture Cup's business plan competition in Norrland, Sweden. In 2014 Calejo became part of the prestigous "33-list", Affärsvärlden´s and Ny Teknik`s list over the 33 most important Swedish technology companies for the future.
What makes the model unique?
The model distinguishes itself by including a large amount of fundamental data and taking into account the frequencies of market fluctuations. This makes it unique on the commodity market.
How accurate is Bright Market Insight?
Bright Market Insight forecasts indicate the correct price directions over 80% of the time for all time horizons longer than 6 months. For comparison purposes, it may be noted that the price is on the rise 58 percent of the time in a 12-month range. The model is also able to identify whether the price will increase or decrease in nearly 80–85 percent of cases where long-range forecasts are made. In a 24-month range, the forecast is very close to the actual price trend. The forecast model achieves its best results in an 18-month range.
How credible are the forecasts?
All results presented are cross-validated, which fully offsets the effects of any overfitting of the results. This means that the set of data is divided up into separate time periods of 1½–2 years each. The results are then simulated in each time period through a model that has been built up solely on the basis of data from the other periods. The graphs presented are combinations of the tests on each individual time period and thus represent the estimated results for entirely new data.
Cross-validation is a very strict test that few statistical models pass. The test is also fully proportional to the difficulty of making forecasts into the future based solely on historical data. We therefore do not expect the results to be overestimated due to overfitting.
How unique is the research behind Bright Market Insight?
Bright Market Insight develops its products in contact with the absolute cutting edge in the field of research and strives to make a break-through in it. The company has developed algorithms surpassing the latest scientific articles and has specifically customized them for its markets. This makes it possible to avoid the problem of selecting variables altogether by allowing networks to select themselves what is relevant and in part also use loosely correlated independent variables, without disrupting the prediction on a large scale.
How does Bright Market Insight work?
Bright Market Insight's model identifies patterns from the historical price trend and its fundamental factors. They are then used to make a price forecast for the future.
The model finds trends in historical data and makes very accurate price trend forecasts in a range of 10–30 months.
Which input data are used?
Bright Market Insight uses tweleve different input data. These are price trend, pulp prices (NBSK and BHKP in NA, Europe and China), producers stocks, shipments of NBSK and BHKP, Chinese imports and exchange rates (Euro/USD). Bright Market Insight Timber uses eight different input data - Swedish and US prices for red- and whitewood, Swedish production, Swedish deliveries, Swedish inventories, housing starts and exchange rates.
Why are not more and other input data used?
We have during the development of the model tested a wide variety of input data, but chosed to retain only those that have a significant impact on the accuracy of forecasts. To keep the model as clean and manageable as possible, we have decided to clear all input data that not have a significant impact on the future prices.
What sets Bright Market Insight from other mathematical models?
All purely statistical models have one primary source of error: that the model is designed to fit the historical trend too specifically instead of finding patterns and being representative of the future. Unfortunately, it is more the rule rather than the exception for economic forecast models that this source of error is not taken up. Oftentimes, their historical accuracy is based entirely on the model fitting the history, and the results thus become totally deceptive.
Bright Market Insight offsets this with a method that automatically prevents it from happening. This "regularization" automatically prevents the model from identifying improbable and far too detailed patterns in rate fluctuations. Together with the flexibility of the neural networks, this provides a very good forecasting ability. Networks without regularization have outperformed competing economic models in almost all scientific studies on real markets.
How is the model's performance secured and evaluated?
To evaluate the model, a very strict test is then conducted: cross-validation. This means that the model will be forced to identify patterns in the price trend from a series of time periods entirely on its own and these patterns are then evaluated on a separate time period. This process is repeated with different combinations of time periods for pattern identification and for evaluation. The results are therefore expected to also be fully representative of future forecasts.
Which are the competitors to Bright Market Insight?
Competing alternatives are based on very simple linear models without regularization or on differential equations, whose forecasting ability lacks a scientific basis, and probably a nonexistent forecasting ability. They have generally not been tested in a structured manner. There are currently no competing forecast models specific to the pulp market at all that are based on more advanced statistical analysis. Copeting companies instead rely on personal market analysts and technical (most likely linear) regression analyses.
What separates the model output common fundamental analysis?
Man has an inherently weak ability to generalize and predict the future. Various scientific studies show that simple statistical methods almost always surpass the predictions or diagnoses of human analysts and experts. There are many reasons for that, one being that analysts often base their own conclusions on simple rules of thumb. These rules of thumb are furthermore rarely or never validated against historical data.
Who is the architect of Bright Market Insight?
Leonard Johard is the architect of the Bright Market Index ANN forecasts and is likewise responsible for the company's product development. Leonard Johard, holds an PhD in perceptual robotics from Scuola Superiore Sant’Anna in Pisa, Italy, and is today a researcher in Ariticial Intelligence at Innopolis University in Kazan, Russia. He has also a background as an energy market analyst at Telge Energi in Sweden.
What does it cost to subscribe to Bright Market Insight?
The prices for Bright Market Insight are as follows:
- Bright Market Insight Light. Bright Market Insight 12 times a year with market analyses and updated pulp forecasts for one regional market 30 months into the future. Price EUR 5,980.
- Bright Market Insight Full. Market analyses and updated pulp forecasts 12 times a year for NA, Europe and Asia with a time horizon of up to 30 months. Published at the beginning of each new month. Price: EUR 11,980.
What benefits can my company get by Bright Market Insight?
The tools that Bright Market Insight can offer is totally unique and revolutionary for the industry.
- Better knowledge of the future creates large inventories and trade profits.
- Better insight and knowledge also help management teams to plan in time for organizational changes, structural measures, purchases and investments.
The savings are substantial.
FAQ: ARTIFICIAL NEURAL NETWORKS (ANN)
The forecasting model is built with the support of ANN. What is ANN?
An artificial neural network, ANN, is a system designed to mimic the ability of brain cells to process its information. Practically all knowledge that people acquire during their lives is created by the brain cells running statistical analyses of information from the senses.
Information processing is decentralized and each neuron works independently to analyze the information in its immediate vicinity, but also assists the rest of the brain cells in processing information through its structure. The result is a very robust statistical analysis, which can handle enormous volumes of information and very complex relationships simultaneously.
How functional are the ANN models?
In various published research studies, neural networks have been compared with linear models, ARIMA models, human analysts etc. These studies clearly demonstrate that neural networks surpass all other known models, especially in long ranges. Neural networks have also proven to be the most successful self-learning algorithm in non-economic applications, with numerous applications in industry and research. For example, the most effective known algorithm for automatic classification of text is a type of ANN.
In what other contexts do we find ANN?
Simple forecast models designed on the basis of ANN are used all the time these days to aid forecasts in the foreign exchange and stock markets, but appear to be rough around the edges from the looks of the public material. For commodity price forecasting, the industry is still forced to rely on human and technical regression analysis. Due to the long-term perspectives, the historical series are not long enough for the use of neural networks without regularization, a unique method that is currently still very rare and no competition is to be expected in the commodity market.
Data strategy is everythingData availability is usually taken for granted, but the data has already…
Calejo Industrial Intelligence ABWith the help of AI and advanced data analysis, Calejo Industrial Intelligence…
Calejo was established in May 2013.Calejo helps research and industry to create and extract the…
Market & Business Intelligence
Market intelligence is the information relevant to a company’s markets, gathered and analyzed specifically for the…
What is Future Intelligence? Future Intelligence predicts decisive shifts in society and technology and registers…
Calejo Health Intelligence ABCalejo Health Intelligence is engaged in the development and sale of services…