Forecasting Brazilian Ethanol Spot Prices Using LSTM

Round 1

Reviewer 1 Report

The paper explores the application of the LSTM network to model and forecast the  Ethanol dynamics. Below are my comments:

1. "The inputs used in the proposed model vary according to the forecast horizon used." More information on the input data should be reported.
2. All the quantities present in equations from 2 to 7 must be defined. You should specify the size of the objects and furthermore, W_fh and all the bias terms have not been defined.
3. It would be helpful if the authors included a brief paragraph introducing neural networks in a general framework before describing the LSTM networks. Especially since the architecture they use includes a dense layer.
4. From figures 7-9 it would appear that a validation set was used during the training of the neural network. This is a good approach to avoid overfitting. The authors should discuss this point in more detail.
5. Section 4.2 illustrates the trend of the loss function during training. It would be logical to discuss this part before the forecasting results as the training phase is done chronologically before.
6. Please, explain the meaning of SVML and SVMR. I guess SVML is the linear SVM and SVMR is the SVM with radial basis function. What values were used for the kernel parameter? Usually the value is chosen using 10-fold cross validation. Please explain in more detail.
7. The robustness of deep learning models is a crucial  point in deep learning applications and it is very discussed in the literature. A short paragraph on the sensitivity  of the LSTM model with respect to hyperparameters could be reported.

Minor points:

• the acronyms such as ARIMA, GARCH CEPEA should be defined.
• “Dense” should be “dense”.
• Formulas 8 and 9 should be written using the notation y_i to denote the observed values and \ hat {y} _i for the predicted values.
• The term “parameters” should be replaced with “measures” when used to describe error measures such as MAPE, RMSE and so on.
• What dropout values were used? Please explain more.
• equation 1: a dot is missing.

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Reviewer 2 Report

Journal: Energies

Article title: Forecasting Brazilian Ethanol Spot Prices using LSTM

General Comments:

This article studies Ethanol Spot Prices in Brazil, which is the second-largest producer of this biofuel in the world. The authors use the neural networks - Long short-term memory (LSTM) - to predict Brazilian ethanol spot prices for three horizon-times (12, 6 and 3 months ahead); the proposed model is compared to three benchmark algorithms: Random Forest, SVM Linear and RBF and evaluate statistical parameters such asMSE (Mean Squared Error), MAPE (Mean Absolute Percentage Error), and accuracy to assess the algorithm robustness. The findings suggest that the LSTM outperforms the other techniques in regression, considering both MSE and MAPE but SVM Linear is better to identify price trends.

Overview:

The paper is well written and the empirical work appears to be carefully and correctly done. The research question is QUITE GOOD and it does make a sufficient new contribution to the literature to be suitable for the ENERGIES. In fact, the literature on influence of the Brazilian Ethanol Spot Prices is quite inexistent. The MAJOR contribution of the paper is the prediction of  Brazilian ethanol spot prices for three horizon-times (12, 6 and 3 months ahead) using neural networks to three benchmark algorithms: Random Forest, SVM Linear and RBF. The paper is very interesting; in my view, it needs to be MINOR improved to reach the standard required for publication in this journal.

Specific Comments:

1. Introduction: theoretical approach + novelty + results (better explanation);
2. Why the authors use the Ethanol Spot Prices in Brazil to prove their ideas?
3. Why the authors investigate ONLY the ethanol and only the Brazil? (EU is NOT the main producer of paper)?
4. Theoretical approach: a paragraph
5. Literature review: try to increase to minimum one page
6. Model: why the presented models are quite wrong regarding the prediction?
7. The COVID-19 increased volatility has not any influence on the models?
8. Conclusions: 1 paragraph with policy implications/limitations
9. References: increase the articles after 2015/ decrease the articles before 2015

General considerations: the idea of the article is very interesting; the results are good and with better respect for literature review, results and explanations (minor changes), it can be published in ENERGIES.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors resolved the issues I had raised. Therefore, I suggest accepting the paper.