BDI & SCFI prediction (‘22/26W)

In the 26th week, BDI and SCFI predicted values and actual values are compared and tested.

---

1.     Last Week’s forecasts

1-1.        BDI, Actual Vs. Predicted

Date	Last week Forecast	Actual BDI	MAPE(%)	Accuracy(%)
‘22/25W(6/20~24)	2,416.467	2,422.80	0.3%	99.7%

---

1-2.        SCFI, Actual Vs. Predicted

Date	Last week Forecast	Actual SCFI	MAPE(%)	Accuracy(%)
‘22/25W(6/20~24)	4,218.104	4,216.13	0.1%	99.9%

 The prediction accuracy of BDI and SCFI was 99.7% and 99.9%, respectively. The predicted value (2,416.467) and actual value (2,422.80) of BDI, and the predicted value (4,218.105) and actual value (4,216.13) of SCFI all achieved accuracy and trend.

 Since the current model includes only univariate, economic variables are excluded. The global economic slowdown seems to be the biggest issue due to the US interest rate hike, but the current model has a limit that cannot be understood. However, since the index itself is a comprehensive index that includes both supply/demand/economy/commodities, if the reliability of the current index forecast continues to be improved, it is expected to be able to effectively respond to economic shocks.

  The BDI average MAPE value for the latest 13 predictions is 4.4% and the prediction rate is 95.6%, and the SCFI average MAPE value is 0.6% and the prediction rate is 99.4%.

---

2.     BDI forecast

2-1. ARIMA, Time series analysis

 For the BDI forecast, I used weekly data (437 pieces) from January 3, 2014 to Jun 24, 2022 to predict the freight for three weeks in the future.

Date	Forecast	Trend(WoW)
‘22/26W(6/27~7/1)	2,527.450	↑
‘22/27W(7/4~8)	2,612.710	↑
‘22/28W(7/11~15)	2,647.274	↗

 As shown in the table and figure above, BDI (as of 6/24: 2,422.80) rose 1.2% from the previous week (as of 6/17: 2,394.20). The overall forecast trend and average are correct, but what is worrying is that the BDI has declined sharply since the middle of last week (Wednesday). The BDI value on Friday was 2,331, which is about 100 points or more different from the average value (2,422.80). However, after the temporary decline, BDI is expected to rise again.

 The BDI prediction result of the ARIMA model is also expected to rise for the next three weeks.

Ref) BDI & SCFI prediction (‘22/25W) (tistory.com)

BDI & SCFI prediction (‘22/25W)

 

2-2. Logistic regression analysis

The logistic regression analysis result was 0.52, which was higher than the baseline 0.5. As it rose significantly from the previous week's forecast (0.34), the logistic analysis model also predicted that the BDI would rise next week.

 

---

3.     SCFI Outlook

3-1. ARIMA, Time series analysis

For the SCFI forecast, I used weekly data (636 pieces) from October 16, 2009 to Jun 24, 2022 to predict the future freight for three weeks.

Date	Forecast	Trend(WoW)
‘22/26W(6/27~7/1)	4,211.002	↘
‘22/27W(7/4~8)	4,210.835	↘
‘22/28W(7/11~15)	4,208.358	↘

 

 As shown in the table and figure above, the SCFI (as of 6/24: 4,216.13) fell -0.14% compared to the previous week (as of 6/17: 4,221.96). Looking at each route, the decline was relatively large in the North American West Coast/East Coast route. The decline in the North American route, which has driven SCFI's uptrend so far, is expected to keep SCFI down going forward.

---

3-2. Logistic regression analysis

  

The predicted value of the logistic regression analysis model of SCFI was 0.386, which predicted the direction of the SCFI freight rate to decrease next week. As the result of the ARIMA model is the same, the SCFI is highly likely to fall.

---

3-3. SVR model

 

 On the other hand, in the SVM model, unlike the previous models (ARIMA, Logistics), the direction of SCFI was predicted to rise.

 Various models are used to objectively forecast freight rates, but it is judged that the predictive power of the SVR model is not high in a short-term approach (within 3 weeks). In particular, in the SCFI freight rate forecast, both qualitatively and quantitatively, freight rates are expected to decline, but in the SVR model, freight rates appear to rise.   After looking at the SCFI movement next week, if the trend does not match, I will exclude the SVR model from future analysis.

 

Date	Forecast	Trend(WoW)
‘22/26W(6/27~7/1)	4,306.766	↗
‘22/27W(7/4~8)	4,359.549	↗
‘22/28W(7/11~15)	4,411.224	↗

Thanks.